fs/btrfs/ioctl.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Copyright (C) 2007 Oracle.  All rights reserved.
   4  */
   5
   6 #include <linux/kernel.h>
   7 #include <linux/bio.h>
   8 #include <linux/file.h>
   9 #include <linux/fs.h>
  10 #include <linux/fsnotify.h>
  11 #include <linux/pagemap.h>
  12 #include <linux/highmem.h>
  13 #include <linux/time.h>
  14 #include <linux/string.h>
  15 #include <linux/backing-dev.h>
  16 #include <linux/mount.h>
  17 #include <linux/namei.h>
  18 #include <linux/writeback.h>
  19 #include <linux/compat.h>
  20 #include <linux/security.h>
  21 #include <linux/xattr.h>
  22 #include <linux/mm.h>
  23 #include <linux/slab.h>
  24 #include <linux/blkdev.h>
  25 #include <linux/uuid.h>
  26 #include <linux/btrfs.h>
  27 #include <linux/uaccess.h>
  28 #include <linux/iversion.h>
  29 #include "ctree.h"
  30 #include "disk-io.h"
  31 #include "export.h"
  32 #include "transaction.h"
  33 #include "btrfs_inode.h"
  34 #include "print-tree.h"
  35 #include "volumes.h"
  36 #include "locking.h"
  37 #include "inode-map.h"
  38 #include "backref.h"
  39 #include "rcu-string.h"
  40 #include "send.h"
  41 #include "dev-replace.h"
  42 #include "props.h"
  43 #include "sysfs.h"
  44 #include "qgroup.h"
  45 #include "tree-log.h"
  46 #include "compression.h"
  47 #include "space-info.h"
  48 #include "delalloc-space.h"
  49 #include "block-group.h"
  50
  51 #ifdef CONFIG_64BIT
  52 /* If we have a 32-bit userspace and 64-bit kernel, then the UAPI
  53  * structures are incorrect, as the timespec structure from userspace
  54  * is 4 bytes too small. We define these alternatives here to teach
  55  * the kernel about the 32-bit struct packing.
  56  */
  57 struct btrfs_ioctl_timespec_32 {
  58         __u64 sec;
  59         __u32 nsec;
  60 } __attribute__ ((__packed__));
  61
  62 struct btrfs_ioctl_received_subvol_args_32 {
  63         char    uuid[BTRFS_UUID_SIZE];  /* in */
  64         __u64   stransid;               /* in */
  65         __u64   rtransid;               /* out */
  66         struct btrfs_ioctl_timespec_32 stime; /* in */
  67         struct btrfs_ioctl_timespec_32 rtime; /* out */
  68         __u64   flags;                  /* in */
  69         __u64   reserved[16];           /* in */
  70 } __attribute__ ((__packed__));
  71
  72 #define BTRFS_IOC_SET_RECEIVED_SUBVOL_32 _IOWR(BTRFS_IOCTL_MAGIC, 37, \
  73                                 struct btrfs_ioctl_received_subvol_args_32)
  74 #endif
  75
  76 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
  77 struct btrfs_ioctl_send_args_32 {
  78         __s64 send_fd;                  /* in */
  79         __u64 clone_sources_count;      /* in */
  80         compat_uptr_t clone_sources;    /* in */
  81         __u64 parent_root;              /* in */
  82         __u64 flags;                    /* in */
  83         __u64 reserved[4];              /* in */
  84 } __attribute__ ((__packed__));
  85
  86 #define BTRFS_IOC_SEND_32 _IOW(BTRFS_IOCTL_MAGIC, 38, \
  87                                struct btrfs_ioctl_send_args_32)
  88 #endif
  89
  90 /* Mask out flags that are inappropriate for the given type of inode. */
  91 static unsigned int btrfs_mask_fsflags_for_type(struct inode *inode,
  92                 unsigned int flags)
  93 {
  94         if (S_ISDIR(inode->i_mode))
  95                 return flags;
  96         else if (S_ISREG(inode->i_mode))
  97                 return flags & ~FS_DIRSYNC_FL;
  98         else
  99                 return flags & (FS_NODUMP_FL | FS_NOATIME_FL);
 100 }
 101
 102 /*
 103  * Export internal inode flags to the format expected by the FS_IOC_GETFLAGS
 104  * ioctl.
 105  */
 106 static unsigned int btrfs_inode_flags_to_fsflags(unsigned int flags)
 107 {
 108         unsigned int iflags = 0;
 109
 110         if (flags & BTRFS_INODE_SYNC)
 111                 iflags |= FS_SYNC_FL;
 112         if (flags & BTRFS_INODE_IMMUTABLE)
 113                 iflags |= FS_IMMUTABLE_FL;
 114         if (flags & BTRFS_INODE_APPEND)
 115                 iflags |= FS_APPEND_FL;
 116         if (flags & BTRFS_INODE_NODUMP)
 117                 iflags |= FS_NODUMP_FL;
 118         if (flags & BTRFS_INODE_NOATIME)
 119                 iflags |= FS_NOATIME_FL;
 120         if (flags & BTRFS_INODE_DIRSYNC)
 121                 iflags |= FS_DIRSYNC_FL;
 122         if (flags & BTRFS_INODE_NODATACOW)
 123                 iflags |= FS_NOCOW_FL;
 124
 125         if (flags & BTRFS_INODE_NOCOMPRESS)
 126                 iflags |= FS_NOCOMP_FL;
 127         else if (flags & BTRFS_INODE_COMPRESS)
 128                 iflags |= FS_COMPR_FL;
 129
 130         return iflags;
 131 }
 132
 133 /*
 134  * Update inode->i_flags based on the btrfs internal flags.
 135  */
 136 void btrfs_sync_inode_flags_to_i_flags(struct inode *inode)
 137 {
 138         struct btrfs_inode *binode = BTRFS_I(inode);
 139         unsigned int new_fl = 0;
 140
 141         if (binode->flags & BTRFS_INODE_SYNC)
 142                 new_fl |= S_SYNC;
 143         if (binode->flags & BTRFS_INODE_IMMUTABLE)
 144                 new_fl |= S_IMMUTABLE;
 145         if (binode->flags & BTRFS_INODE_APPEND)
 146                 new_fl |= S_APPEND;
 147         if (binode->flags & BTRFS_INODE_NOATIME)
 148                 new_fl |= S_NOATIME;
 149         if (binode->flags & BTRFS_INODE_DIRSYNC)
 150                 new_fl |= S_DIRSYNC;
 151
 152         set_mask_bits(&inode->i_flags,
 153                       S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME | S_DIRSYNC,
 154                       new_fl);
 155 }
 156
 157 static int btrfs_ioctl_getflags(struct file *file, void __user *arg)
 158 {
 159         struct btrfs_inode *binode = BTRFS_I(file_inode(file));
 160         unsigned int flags = btrfs_inode_flags_to_fsflags(binode->flags);
 161
 162         if (copy_to_user(arg, &flags, sizeof(flags)))
 163                 return -EFAULT;
 164         return 0;
 165 }
 166
 167 /*
 168  * Check if @flags are a supported and valid set of FS_*_FL flags and that
 169  * the old and new flags are not conflicting
 170  */
 171 static int check_fsflags(unsigned int old_flags, unsigned int flags)
 172 {
 173         if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \
 174                       FS_NOATIME_FL | FS_NODUMP_FL | \
 175                       FS_SYNC_FL | FS_DIRSYNC_FL | \
 176                       FS_NOCOMP_FL | FS_COMPR_FL |
 177                       FS_NOCOW_FL))
 178                 return -EOPNOTSUPP;
 179
 180         /* COMPR and NOCOMP on new/old are valid */
 181         if ((flags & FS_NOCOMP_FL) && (flags & FS_COMPR_FL))
 182                 return -EINVAL;
 183
 184         if ((flags & FS_COMPR_FL) && (flags & FS_NOCOW_FL))
 185                 return -EINVAL;
 186
 187         /* NOCOW and compression options are mutually exclusive */
 188         if ((old_flags & FS_NOCOW_FL) && (flags & (FS_COMPR_FL | FS_NOCOMP_FL)))
 189                 return -EINVAL;
 190         if ((flags & FS_NOCOW_FL) && (old_flags & (FS_COMPR_FL | FS_NOCOMP_FL)))
 191                 return -EINVAL;
 192
 193         return 0;
 194 }
 195
 196 static int btrfs_ioctl_setflags(struct file *file, void __user *arg)
 197 {
 198         struct inode *inode = file_inode(file);
 199         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 200         struct btrfs_inode *binode = BTRFS_I(inode);
 201         struct btrfs_root *root = binode->root;
 202         struct btrfs_trans_handle *trans;
 203         unsigned int fsflags, old_fsflags;
 204         int ret;
 205         const char *comp = NULL;
 206         u32 binode_flags;
 207
 208         if (!inode_owner_or_capable(inode))
 209                 return -EPERM;
 210
 211         if (btrfs_root_readonly(root))
 212                 return -EROFS;
 213
 214         if (copy_from_user(&fsflags, arg, sizeof(fsflags)))
 215                 return -EFAULT;
 216
 217         ret = mnt_want_write_file(file);
 218         if (ret)
 219                 return ret;
 220
 221         inode_lock(inode);
 222         fsflags = btrfs_mask_fsflags_for_type(inode, fsflags);
 223         old_fsflags = btrfs_inode_flags_to_fsflags(binode->flags);
 224
 225         ret = vfs_ioc_setflags_prepare(inode, old_fsflags, fsflags);
 226         if (ret)
 227                 goto out_unlock;
 228
 229         ret = check_fsflags(old_fsflags, fsflags);
 230         if (ret)
 231                 goto out_unlock;
 232
 233         binode_flags = binode->flags;
 234         if (fsflags & FS_SYNC_FL)
 235                 binode_flags |= BTRFS_INODE_SYNC;
 236         else
 237                 binode_flags &= ~BTRFS_INODE_SYNC;
 238         if (fsflags & FS_IMMUTABLE_FL)
 239                 binode_flags |= BTRFS_INODE_IMMUTABLE;
 240         else
 241                 binode_flags &= ~BTRFS_INODE_IMMUTABLE;
 242         if (fsflags & FS_APPEND_FL)
 243                 binode_flags |= BTRFS_INODE_APPEND;
 244         else
 245                 binode_flags &= ~BTRFS_INODE_APPEND;
 246         if (fsflags & FS_NODUMP_FL)
 247                 binode_flags |= BTRFS_INODE_NODUMP;
 248         else
 249                 binode_flags &= ~BTRFS_INODE_NODUMP;
 250         if (fsflags & FS_NOATIME_FL)
 251                 binode_flags |= BTRFS_INODE_NOATIME;
 252         else
 253                 binode_flags &= ~BTRFS_INODE_NOATIME;
 254         if (fsflags & FS_DIRSYNC_FL)
 255                 binode_flags |= BTRFS_INODE_DIRSYNC;
 256         else
 257                 binode_flags &= ~BTRFS_INODE_DIRSYNC;
 258         if (fsflags & FS_NOCOW_FL) {
 259                 if (S_ISREG(inode->i_mode)) {
 260                         /*
 261                          * It's safe to turn csums off here, no extents exist.
 262                          * Otherwise we want the flag to reflect the real COW
 263                          * status of the file and will not set it.
 264                          */
 265                         if (inode->i_size == 0)
 266                                 binode_flags |= BTRFS_INODE_NODATACOW |
 267                                                 BTRFS_INODE_NODATASUM;
 268                 } else {
 269                         binode_flags |= BTRFS_INODE_NODATACOW;
 270                 }
 271         } else {
 272                 /*
 273                  * Revert back under same assumptions as above
 274                  */
 275                 if (S_ISREG(inode->i_mode)) {
 276                         if (inode->i_size == 0)
 277                                 binode_flags &= ~(BTRFS_INODE_NODATACOW |
 278                                                   BTRFS_INODE_NODATASUM);
 279                 } else {
 280                         binode_flags &= ~BTRFS_INODE_NODATACOW;
 281                 }
 282         }
 283
 284         /*
 285          * The COMPRESS flag can only be changed by users, while the NOCOMPRESS
 286          * flag may be changed automatically if compression code won't make
 287          * things smaller.
 288          */
 289         if (fsflags & FS_NOCOMP_FL) {
 290                 binode_flags &= ~BTRFS_INODE_COMPRESS;
 291                 binode_flags |= BTRFS_INODE_NOCOMPRESS;
 292         } else if (fsflags & FS_COMPR_FL) {
 293
 294                 if (IS_SWAPFILE(inode)) {
 295                         ret = -ETXTBSY;
 296                         goto out_unlock;
 297                 }
 298
 299                 binode_flags |= BTRFS_INODE_COMPRESS;
 300                 binode_flags &= ~BTRFS_INODE_NOCOMPRESS;
 301
 302                 comp = btrfs_compress_type2str(fs_info->compress_type);
 303                 if (!comp || comp[0] == 0)
 304                         comp = btrfs_compress_type2str(BTRFS_COMPRESS_ZLIB);
 305         } else {
 306                 binode_flags &= ~(BTRFS_INODE_COMPRESS | BTRFS_INODE_NOCOMPRESS);
 307         }
 308
 309         /*
 310          * 1 for inode item
 311          * 2 for properties
 312          */
 313         trans = btrfs_start_transaction(root, 3);
 314         if (IS_ERR(trans)) {
 315                 ret = PTR_ERR(trans);
 316                 goto out_unlock;
 317         }
 318
 319         if (comp) {
 320                 ret = btrfs_set_prop(trans, inode, "btrfs.compression", comp,
 321                                      strlen(comp), 0);
 322                 if (ret) {
 323                         btrfs_abort_transaction(trans, ret);
 324                         goto out_end_trans;
 325                 }
 326         } else {
 327                 ret = btrfs_set_prop(trans, inode, "btrfs.compression", NULL,
 328                                      0, 0);
 329                 if (ret && ret != -ENODATA) {
 330                         btrfs_abort_transaction(trans, ret);
 331                         goto out_end_trans;
 332                 }
 333         }
 334
 335         binode->flags = binode_flags;
 336         btrfs_sync_inode_flags_to_i_flags(inode);
 337         inode_inc_iversion(inode);
 338         inode->i_ctime = current_time(inode);
 339         ret = btrfs_update_inode(trans, root, inode);
 340
 341  out_end_trans:
 342         btrfs_end_transaction(trans);
 343  out_unlock:
 344         inode_unlock(inode);
 345         mnt_drop_write_file(file);
 346         return ret;
 347 }
 348
 349 /*
 350  * Translate btrfs internal inode flags to xflags as expected by the
 351  * FS_IOC_FSGETXATT ioctl. Filter only the supported ones, unknown flags are
 352  * silently dropped.
 353  */
 354 static unsigned int btrfs_inode_flags_to_xflags(unsigned int flags)
 355 {
 356         unsigned int xflags = 0;
 357
 358         if (flags & BTRFS_INODE_APPEND)
 359                 xflags |= FS_XFLAG_APPEND;
 360         if (flags & BTRFS_INODE_IMMUTABLE)
 361                 xflags |= FS_XFLAG_IMMUTABLE;
 362         if (flags & BTRFS_INODE_NOATIME)
 363                 xflags |= FS_XFLAG_NOATIME;
 364         if (flags & BTRFS_INODE_NODUMP)
 365                 xflags |= FS_XFLAG_NODUMP;
 366         if (flags & BTRFS_INODE_SYNC)
 367                 xflags |= FS_XFLAG_SYNC;
 368
 369         return xflags;
 370 }
 371
 372 /* Check if @flags are a supported and valid set of FS_XFLAGS_* flags */
 373 static int check_xflags(unsigned int flags)
 374 {
 375         if (flags & ~(FS_XFLAG_APPEND | FS_XFLAG_IMMUTABLE | FS_XFLAG_NOATIME |
 376                       FS_XFLAG_NODUMP | FS_XFLAG_SYNC))
 377                 return -EOPNOTSUPP;
 378         return 0;
 379 }
 380
 381 /*
 382  * Set the xflags from the internal inode flags. The remaining items of fsxattr
 383  * are zeroed.
 384  */
 385 static int btrfs_ioctl_fsgetxattr(struct file *file, void __user *arg)
 386 {
 387         struct btrfs_inode *binode = BTRFS_I(file_inode(file));
 388         struct fsxattr fa;
 389
 390         simple_fill_fsxattr(&fa, btrfs_inode_flags_to_xflags(binode->flags));
 391         if (copy_to_user(arg, &fa, sizeof(fa)))
 392                 return -EFAULT;
 393
 394         return 0;
 395 }
 396
 397 static int btrfs_ioctl_fssetxattr(struct file *file, void __user *arg)
 398 {
 399         struct inode *inode = file_inode(file);
 400         struct btrfs_inode *binode = BTRFS_I(inode);
 401         struct btrfs_root *root = binode->root;
 402         struct btrfs_trans_handle *trans;
 403         struct fsxattr fa, old_fa;
 404         unsigned old_flags;
 405         unsigned old_i_flags;
 406         int ret = 0;
 407
 408         if (!inode_owner_or_capable(inode))
 409                 return -EPERM;
 410
 411         if (btrfs_root_readonly(root))
 412                 return -EROFS;
 413
 414         if (copy_from_user(&fa, arg, sizeof(fa)))
 415                 return -EFAULT;
 416
 417         ret = check_xflags(fa.fsx_xflags);
 418         if (ret)
 419                 return ret;
 420
 421         if (fa.fsx_extsize != 0 || fa.fsx_projid != 0 || fa.fsx_cowextsize != 0)
 422                 return -EOPNOTSUPP;
 423
 424         ret = mnt_want_write_file(file);
 425         if (ret)
 426                 return ret;
 427
 428         inode_lock(inode);
 429
 430         old_flags = binode->flags;
 431         old_i_flags = inode->i_flags;
 432
 433         simple_fill_fsxattr(&old_fa,
 434                             btrfs_inode_flags_to_xflags(binode->flags));
 435         ret = vfs_ioc_fssetxattr_check(inode, &old_fa, &fa);
 436         if (ret)
 437                 goto out_unlock;
 438
 439         if (fa.fsx_xflags & FS_XFLAG_SYNC)
 440                 binode->flags |= BTRFS_INODE_SYNC;
 441         else
 442                 binode->flags &= ~BTRFS_INODE_SYNC;
 443         if (fa.fsx_xflags & FS_XFLAG_IMMUTABLE)
 444                 binode->flags |= BTRFS_INODE_IMMUTABLE;
 445         else
 446                 binode->flags &= ~BTRFS_INODE_IMMUTABLE;
 447         if (fa.fsx_xflags & FS_XFLAG_APPEND)
 448                 binode->flags |= BTRFS_INODE_APPEND;
 449         else
 450                 binode->flags &= ~BTRFS_INODE_APPEND;
 451         if (fa.fsx_xflags & FS_XFLAG_NODUMP)
 452                 binode->flags |= BTRFS_INODE_NODUMP;
 453         else
 454                 binode->flags &= ~BTRFS_INODE_NODUMP;
 455         if (fa.fsx_xflags & FS_XFLAG_NOATIME)
 456                 binode->flags |= BTRFS_INODE_NOATIME;
 457         else
 458                 binode->flags &= ~BTRFS_INODE_NOATIME;
 459
 460         /* 1 item for the inode */
 461         trans = btrfs_start_transaction(root, 1);
 462         if (IS_ERR(trans)) {
 463                 ret = PTR_ERR(trans);
 464                 goto out_unlock;
 465         }
 466
 467         btrfs_sync_inode_flags_to_i_flags(inode);
 468         inode_inc_iversion(inode);
 469         inode->i_ctime = current_time(inode);
 470         ret = btrfs_update_inode(trans, root, inode);
 471
 472         btrfs_end_transaction(trans);
 473
 474 out_unlock:
 475         if (ret) {
 476                 binode->flags = old_flags;
 477                 inode->i_flags = old_i_flags;
 478         }
 479
 480         inode_unlock(inode);
 481         mnt_drop_write_file(file);
 482
 483         return ret;
 484 }
 485
 486 static int btrfs_ioctl_getversion(struct file *file, int __user *arg)
 487 {
 488         struct inode *inode = file_inode(file);
 489
 490         return put_user(inode->i_generation, arg);
 491 }
 492
 493 static noinline int btrfs_ioctl_fitrim(struct btrfs_fs_info *fs_info,
 494                                         void __user *arg)
 495 {
 496         struct btrfs_device *device;
 497         struct request_queue *q;
 498         struct fstrim_range range;
 499         u64 minlen = ULLONG_MAX;
 500         u64 num_devices = 0;
 501         int ret;
 502
 503         if (!capable(CAP_SYS_ADMIN))
 504                 return -EPERM;
 505
 506         /*
 507          * If the fs is mounted with nologreplay, which requires it to be
 508          * mounted in RO mode as well, we can not allow discard on free space
 509          * inside block groups, because log trees refer to extents that are not
 510          * pinned in a block group's free space cache (pinning the extents is
 511          * precisely the first phase of replaying a log tree).
 512          */
 513         if (btrfs_test_opt(fs_info, NOLOGREPLAY))
 514                 return -EROFS;
 515
 516         rcu_read_lock();
 517         list_for_each_entry_rcu(device, &fs_info->fs_devices->devices,
 518                                 dev_list) {
 519                 if (!device->bdev)
 520                         continue;
 521                 q = bdev_get_queue(device->bdev);
 522                 if (blk_queue_discard(q)) {
 523                         num_devices++;
 524                         minlen = min_t(u64, q->limits.discard_granularity,
 525                                      minlen);
 526                 }
 527         }
 528         rcu_read_unlock();
 529
 530         if (!num_devices)
 531                 return -EOPNOTSUPP;
 532         if (copy_from_user(&range, arg, sizeof(range)))
 533                 return -EFAULT;
 534
 535         /*
 536          * NOTE: Don't truncate the range using super->total_bytes.  Bytenr of
 537          * block group is in the logical address space, which can be any
 538          * sectorsize aligned bytenr in  the range [0, U64_MAX].
 539          */
 540         if (range.len < fs_info->sb->s_blocksize)
 541                 return -EINVAL;
 542
 543         range.minlen = max(range.minlen, minlen);
 544         ret = btrfs_trim_fs(fs_info, &range);
 545         if (ret < 0)
 546                 return ret;
 547
 548         if (copy_to_user(arg, &range, sizeof(range)))
 549                 return -EFAULT;
 550
 551         return 0;
 552 }
 553
 554 int __pure btrfs_is_empty_uuid(u8 *uuid)
 555 {
 556         int i;
 557
 558         for (i = 0; i < BTRFS_UUID_SIZE; i++) {
 559                 if (uuid[i])
 560                         return 0;
 561         }
 562         return 1;
 563 }
 564
 565 static noinline int create_subvol(struct inode *dir,
 566                                   struct dentry *dentry,
 567                                   const char *name, int namelen,
 568                                   struct btrfs_qgroup_inherit *inherit)
 569 {
 570         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 571         struct btrfs_trans_handle *trans;
 572         struct btrfs_key key;
 573         struct btrfs_root_item *root_item;
 574         struct btrfs_inode_item *inode_item;
 575         struct extent_buffer *leaf;
 576         struct btrfs_root *root = BTRFS_I(dir)->root;
 577         struct btrfs_root *new_root;
 578         struct btrfs_block_rsv block_rsv;
 579         struct timespec64 cur_time = current_time(dir);
 580         struct inode *inode;
 581         int ret;
 582         int err;
 583         dev_t anon_dev = 0;
 584         u64 objectid;
 585         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
 586         u64 index = 0;
 587
 588         root_item = kzalloc(sizeof(*root_item), GFP_KERNEL);
 589         if (!root_item)
 590                 return -ENOMEM;
 591
 592         ret = btrfs_find_free_objectid(fs_info->tree_root, &objectid);
 593         if (ret)
 594                 goto fail_free;
 595
 596         ret = get_anon_bdev(&anon_dev);
 597         if (ret < 0)
 598                 goto fail_free;
 599
 600         /*
 601          * Don't create subvolume whose level is not zero. Or qgroup will be
 602          * screwed up since it assumes subvolume qgroup's level to be 0.
 603          */
 604         if (btrfs_qgroup_level(objectid)) {
 605                 ret = -ENOSPC;
 606                 goto fail_free;
 607         }
 608
 609         btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
 610         /*
 611          * The same as the snapshot creation, please see the comment
 612          * of create_snapshot().
 613          */
 614         ret = btrfs_subvolume_reserve_metadata(root, &block_rsv, 8, false);
 615         if (ret)
 616                 goto fail_free;
 617
 618         trans = btrfs_start_transaction(root, 0);
 619         if (IS_ERR(trans)) {
 620                 ret = PTR_ERR(trans);
 621                 btrfs_subvolume_release_metadata(fs_info, &block_rsv);
 622                 goto fail_free;
 623         }
 624         trans->block_rsv = &block_rsv;
 625         trans->bytes_reserved = block_rsv.size;
 626
 627         ret = btrfs_qgroup_inherit(trans, 0, objectid, inherit);
 628         if (ret)
 629                 goto fail;
 630
 631         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
 632         if (IS_ERR(leaf)) {
 633                 ret = PTR_ERR(leaf);
 634                 goto fail;
 635         }
 636
 637         btrfs_mark_buffer_dirty(leaf);
 638
 639         inode_item = &root_item->inode;
 640         btrfs_set_stack_inode_generation(inode_item, 1);
 641         btrfs_set_stack_inode_size(inode_item, 3);
 642         btrfs_set_stack_inode_nlink(inode_item, 1);
 643         btrfs_set_stack_inode_nbytes(inode_item,
 644                                      fs_info->nodesize);
 645         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
 646
 647         btrfs_set_root_flags(root_item, 0);
 648         btrfs_set_root_limit(root_item, 0);
 649         btrfs_set_stack_inode_flags(inode_item, BTRFS_INODE_ROOT_ITEM_INIT);
 650
 651         btrfs_set_root_bytenr(root_item, leaf->start);
 652         btrfs_set_root_generation(root_item, trans->transid);
 653         btrfs_set_root_level(root_item, 0);
 654         btrfs_set_root_refs(root_item, 1);
 655         btrfs_set_root_used(root_item, leaf->len);
 656         btrfs_set_root_last_snapshot(root_item, 0);
 657
 658         btrfs_set_root_generation_v2(root_item,
 659                         btrfs_root_generation(root_item));
 660         generate_random_guid(root_item->uuid);
 661         btrfs_set_stack_timespec_sec(&root_item->otime, cur_time.tv_sec);
 662         btrfs_set_stack_timespec_nsec(&root_item->otime, cur_time.tv_nsec);
 663         root_item->ctime = root_item->otime;
 664         btrfs_set_root_ctransid(root_item, trans->transid);
 665         btrfs_set_root_otransid(root_item, trans->transid);
 666
 667         btrfs_tree_unlock(leaf);
 668         free_extent_buffer(leaf);
 669         leaf = NULL;
 670
 671         btrfs_set_root_dirid(root_item, new_dirid);
 672
 673         key.objectid = objectid;
 674         key.offset = 0;
 675         key.type = BTRFS_ROOT_ITEM_KEY;
 676         ret = btrfs_insert_root(trans, fs_info->tree_root, &key,
 677                                 root_item);
 678         if (ret)
 679                 goto fail;
 680
 681         key.offset = (u64)-1;
 682         new_root = btrfs_get_new_fs_root(fs_info, objectid, anon_dev);
 683         if (IS_ERR(new_root)) {
 684                 free_anon_bdev(anon_dev);
 685                 ret = PTR_ERR(new_root);
 686                 btrfs_abort_transaction(trans, ret);
 687                 goto fail;
 688         }
 689         /* Freeing will be done in btrfs_put_root() of new_root */
 690         anon_dev = 0;
 691
 692         btrfs_record_root_in_trans(trans, new_root);
 693
 694         ret = btrfs_create_subvol_root(trans, new_root, root, new_dirid);
 695         btrfs_put_root(new_root);
 696         if (ret) {
 697                 /* We potentially lose an unused inode item here */
 698                 btrfs_abort_transaction(trans, ret);
 699                 goto fail;
 700         }
 701
 702         mutex_lock(&new_root->objectid_mutex);
 703         new_root->highest_objectid = new_dirid;
 704         mutex_unlock(&new_root->objectid_mutex);
 705
 706         /*
 707          * insert the directory item
 708          */
 709         ret = btrfs_set_inode_index(BTRFS_I(dir), &index);
 710         if (ret) {
 711                 btrfs_abort_transaction(trans, ret);
 712                 goto fail;
 713         }
 714
 715         ret = btrfs_insert_dir_item(trans, name, namelen, BTRFS_I(dir), &key,
 716                                     BTRFS_FT_DIR, index);
 717         if (ret) {
 718                 btrfs_abort_transaction(trans, ret);
 719                 goto fail;
 720         }
 721
 722         btrfs_i_size_write(BTRFS_I(dir), dir->i_size + namelen * 2);
 723         ret = btrfs_update_inode(trans, root, dir);
 724         if (ret) {
 725                 btrfs_abort_transaction(trans, ret);
 726                 goto fail;
 727         }
 728
 729         ret = btrfs_add_root_ref(trans, objectid, root->root_key.objectid,
 730                                  btrfs_ino(BTRFS_I(dir)), index, name, namelen);
 731         if (ret) {
 732                 btrfs_abort_transaction(trans, ret);
 733                 goto fail;
 734         }
 735
 736         ret = btrfs_uuid_tree_add(trans, root_item->uuid,
 737                                   BTRFS_UUID_KEY_SUBVOL, objectid);
 738         if (ret)
 739                 btrfs_abort_transaction(trans, ret);
 740
 741 fail:
 742         kfree(root_item);
 743         trans->block_rsv = NULL;
 744         trans->bytes_reserved = 0;
 745         btrfs_subvolume_release_metadata(fs_info, &block_rsv);
 746
 747         err = btrfs_commit_transaction(trans);
 748         if (err && !ret)
 749                 ret = err;
 750
 751         if (!ret) {
 752                 inode = btrfs_lookup_dentry(dir, dentry);
 753                 if (IS_ERR(inode))
 754                         return PTR_ERR(inode);
 755                 d_instantiate(dentry, inode);
 756         }
 757         return ret;
 758
 759 fail_free:
 760         if (anon_dev)
 761                 free_anon_bdev(anon_dev);
 762         kfree(root_item);
 763         return ret;
 764 }
 765
 766 static int create_snapshot(struct btrfs_root *root, struct inode *dir,
 767                            struct dentry *dentry, bool readonly,
 768                            struct btrfs_qgroup_inherit *inherit)
 769 {
 770         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 771         struct inode *inode;
 772         struct btrfs_pending_snapshot *pending_snapshot;
 773         struct btrfs_trans_handle *trans;
 774         int ret;
 775
 776         if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
 777                 return -EINVAL;
 778
 779         if (atomic_read(&root->nr_swapfiles)) {
 780                 btrfs_warn(fs_info,
 781                            "cannot snapshot subvolume with active swapfile");
 782                 return -ETXTBSY;
 783         }
 784
 785         pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_KERNEL);
 786         if (!pending_snapshot)
 787                 return -ENOMEM;
 788
 789         ret = get_anon_bdev(&pending_snapshot->anon_dev);
 790         if (ret < 0)
 791                 goto free_pending;
 792         pending_snapshot->root_item = kzalloc(sizeof(struct btrfs_root_item),
 793                         GFP_KERNEL);
 794         pending_snapshot->path = btrfs_alloc_path();
 795         if (!pending_snapshot->root_item || !pending_snapshot->path) {
 796                 ret = -ENOMEM;
 797                 goto free_pending;
 798         }
 799
 800         btrfs_init_block_rsv(&pending_snapshot->block_rsv,
 801                              BTRFS_BLOCK_RSV_TEMP);
 802         /*
 803          * 1 - parent dir inode
 804          * 2 - dir entries
 805          * 1 - root item
 806          * 2 - root ref/backref
 807          * 1 - root of snapshot
 808          * 1 - UUID item
 809          */
 810         ret = btrfs_subvolume_reserve_metadata(BTRFS_I(dir)->root,
 811                                         &pending_snapshot->block_rsv, 8,
 812                                         false);
 813         if (ret)
 814                 goto free_pending;
 815
 816         pending_snapshot->dentry = dentry;
 817         pending_snapshot->root = root;
 818         pending_snapshot->readonly = readonly;
 819         pending_snapshot->dir = dir;
 820         pending_snapshot->inherit = inherit;
 821
 822         trans = btrfs_start_transaction(root, 0);
 823         if (IS_ERR(trans)) {
 824                 ret = PTR_ERR(trans);
 825                 goto fail;
 826         }
 827
 828         spin_lock(&fs_info->trans_lock);
 829         list_add(&pending_snapshot->list,
 830                  &trans->transaction->pending_snapshots);
 831         spin_unlock(&fs_info->trans_lock);
 832
 833         ret = btrfs_commit_transaction(trans);
 834         if (ret)
 835                 goto fail;
 836
 837         ret = pending_snapshot->error;
 838         if (ret)
 839                 goto fail;
 840
 841         ret = btrfs_orphan_cleanup(pending_snapshot->snap);
 842         if (ret)
 843                 goto fail;
 844
 845         inode = btrfs_lookup_dentry(d_inode(dentry->d_parent), dentry);
 846         if (IS_ERR(inode)) {
 847                 ret = PTR_ERR(inode);
 848                 goto fail;
 849         }
 850
 851         d_instantiate(dentry, inode);
 852         ret = 0;
 853         pending_snapshot->anon_dev = 0;
 854 fail:
 855         /* Prevent double freeing of anon_dev */
 856         if (ret && pending_snapshot->snap)
 857                 pending_snapshot->snap->anon_dev = 0;
 858         btrfs_put_root(pending_snapshot->snap);
 859         btrfs_subvolume_release_metadata(fs_info, &pending_snapshot->block_rsv);
 860 free_pending:
 861         if (pending_snapshot->anon_dev)
 862                 free_anon_bdev(pending_snapshot->anon_dev);
 863         kfree(pending_snapshot->root_item);
 864         btrfs_free_path(pending_snapshot->path);
 865         kfree(pending_snapshot);
 866
 867         return ret;
 868 }
 869
 870 /*  copy of may_delete in fs/namei.c()
 871  *      Check whether we can remove a link victim from directory dir, check
 872  *  whether the type of victim is right.
 873  *  1. We can't do it if dir is read-only (done in permission())
 874  *  2. We should have write and exec permissions on dir
 875  *  3. We can't remove anything from append-only dir
 876  *  4. We can't do anything with immutable dir (done in permission())
 877  *  5. If the sticky bit on dir is set we should either
 878  *      a. be owner of dir, or
 879  *      b. be owner of victim, or
 880  *      c. have CAP_FOWNER capability
 881  *  6. If the victim is append-only or immutable we can't do anything with
 882  *     links pointing to it.
 883  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
 884  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
 885  *  9. We can't remove a root or mountpoint.
 886  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
 887  *     nfs_async_unlink().
 888  */
 889
 890 static int btrfs_may_delete(struct inode *dir, struct dentry *victim, int isdir)
 891 {
 892         int error;
 893
 894         if (d_really_is_negative(victim))
 895                 return -ENOENT;
 896
 897         BUG_ON(d_inode(victim->d_parent) != dir);
 898         audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
 899
 900         error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
 901         if (error)
 902                 return error;
 903         if (IS_APPEND(dir))
 904                 return -EPERM;
 905         if (check_sticky(dir, d_inode(victim)) || IS_APPEND(d_inode(victim)) ||
 906             IS_IMMUTABLE(d_inode(victim)) || IS_SWAPFILE(d_inode(victim)))
 907                 return -EPERM;
 908         if (isdir) {
 909                 if (!d_is_dir(victim))
 910                         return -ENOTDIR;
 911                 if (IS_ROOT(victim))
 912                         return -EBUSY;
 913         } else if (d_is_dir(victim))
 914                 return -EISDIR;
 915         if (IS_DEADDIR(dir))
 916                 return -ENOENT;
 917         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
 918                 return -EBUSY;
 919         return 0;
 920 }
 921
 922 /* copy of may_create in fs/namei.c() */
 923 static inline int btrfs_may_create(struct inode *dir, struct dentry *child)
 924 {
 925         if (d_really_is_positive(child))
 926                 return -EEXIST;
 927         if (IS_DEADDIR(dir))
 928                 return -ENOENT;
 929         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
 930 }
 931
 932 /*
 933  * Create a new subvolume below @parent.  This is largely modeled after
 934  * sys_mkdirat and vfs_mkdir, but we only do a single component lookup
 935  * inside this filesystem so it's quite a bit simpler.
 936  */
 937 static noinline int btrfs_mksubvol(const struct path *parent,
 938                                    const char *name, int namelen,
 939                                    struct btrfs_root *snap_src,
 940                                    bool readonly,
 941                                    struct btrfs_qgroup_inherit *inherit)
 942 {
 943         struct inode *dir = d_inode(parent->dentry);
 944         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
 945         struct dentry *dentry;
 946         int error;
 947
 948         error = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
 949         if (error == -EINTR)
 950                 return error;
 951
 952         dentry = lookup_one_len(name, parent->dentry, namelen);
 953         error = PTR_ERR(dentry);
 954         if (IS_ERR(dentry))
 955                 goto out_unlock;
 956
 957         error = btrfs_may_create(dir, dentry);
 958         if (error)
 959                 goto out_dput;
 960
 961         /*
 962          * even if this name doesn't exist, we may get hash collisions.
 963          * check for them now when we can safely fail
 964          */
 965         error = btrfs_check_dir_item_collision(BTRFS_I(dir)->root,
 966                                                dir->i_ino, name,
 967                                                namelen);
 968         if (error)
 969                 goto out_dput;
 970
 971         down_read(&fs_info->subvol_sem);
 972
 973         if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0)
 974                 goto out_up_read;
 975
 976         if (snap_src)
 977                 error = create_snapshot(snap_src, dir, dentry, readonly, inherit);
 978         else
 979                 error = create_subvol(dir, dentry, name, namelen, inherit);
 980
 981         if (!error)
 982                 fsnotify_mkdir(dir, dentry);
 983 out_up_read:
 984         up_read(&fs_info->subvol_sem);
 985 out_dput:
 986         dput(dentry);
 987 out_unlock:
 988         inode_unlock(dir);
 989         return error;
 990 }
 991
 992 static noinline int btrfs_mksnapshot(const struct path *parent,
 993                                    const char *name, int namelen,
 994                                    struct btrfs_root *root,
 995                                    bool readonly,
 996                                    struct btrfs_qgroup_inherit *inherit)
 997 {
 998         int ret;
 999         bool snapshot_force_cow = false;
1000
1001         /*
1002          * Force new buffered writes to reserve space even when NOCOW is
1003          * possible. This is to avoid later writeback (running dealloc) to
1004          * fallback to COW mode and unexpectedly fail with ENOSPC.
1005          */
1006         btrfs_drew_read_lock(&root->snapshot_lock);
1007
1008         ret = btrfs_start_delalloc_snapshot(root);
1009         if (ret)
1010                 goto out;
1011
1012         /*
1013          * All previous writes have started writeback in NOCOW mode, so now
1014          * we force future writes to fallback to COW mode during snapshot
1015          * creation.
1016          */
1017         atomic_inc(&root->snapshot_force_cow);
1018         snapshot_force_cow = true;
1019
1020         btrfs_wait_ordered_extents(root, U64_MAX, 0, (u64)-1);
1021
1022         ret = btrfs_mksubvol(parent, name, namelen,
1023                              root, readonly, inherit);
1024 out:
1025         if (snapshot_force_cow)
1026                 atomic_dec(&root->snapshot_force_cow);
1027         btrfs_drew_read_unlock(&root->snapshot_lock);
1028         return ret;
1029 }
1030
1031 /*
1032  * When we're defragging a range, we don't want to kick it off again
1033  * if it is really just waiting for delalloc to send it down.
1034  * If we find a nice big extent or delalloc range for the bytes in the
1035  * file you want to defrag, we return 0 to let you know to skip this
1036  * part of the file
1037  */
1038 static int check_defrag_in_cache(struct inode *inode, u64 offset, u32 thresh)
1039 {
1040         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1041         struct extent_map *em = NULL;
1042         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1043         u64 end;
1044
1045         read_lock(&em_tree->lock);
1046         em = lookup_extent_mapping(em_tree, offset, PAGE_SIZE);
1047         read_unlock(&em_tree->lock);
1048
1049         if (em) {
1050                 end = extent_map_end(em);
1051                 free_extent_map(em);
1052                 if (end - offset > thresh)
1053                         return 0;
1054         }
1055         /* if we already have a nice delalloc here, just stop */
1056         thresh /= 2;
1057         end = count_range_bits(io_tree, &offset, offset + thresh,
1058                                thresh, EXTENT_DELALLOC, 1);
1059         if (end >= thresh)
1060                 return 0;
1061         return 1;
1062 }
1063
1064 /*
1065  * helper function to walk through a file and find extents
1066  * newer than a specific transid, and smaller than thresh.
1067  *
1068  * This is used by the defragging code to find new and small
1069  * extents
1070  */
1071 static int find_new_extents(struct btrfs_root *root,
1072                             struct inode *inode, u64 newer_than,
1073                             u64 *off, u32 thresh)
1074 {
1075         struct btrfs_path *path;
1076         struct btrfs_key min_key;
1077         struct extent_buffer *leaf;
1078         struct btrfs_file_extent_item *extent;
1079         int type;
1080         int ret;
1081         u64 ino = btrfs_ino(BTRFS_I(inode));
1082
1083         path = btrfs_alloc_path();
1084         if (!path)
1085                 return -ENOMEM;
1086
1087         min_key.objectid = ino;
1088         min_key.type = BTRFS_EXTENT_DATA_KEY;
1089         min_key.offset = *off;
1090
1091         while (1) {
1092                 ret = btrfs_search_forward(root, &min_key, path, newer_than);
1093                 if (ret != 0)
1094                         goto none;
1095 process_slot:
1096                 if (min_key.objectid != ino)
1097                         goto none;
1098                 if (min_key.type != BTRFS_EXTENT_DATA_KEY)
1099                         goto none;
1100
1101                 leaf = path->nodes[0];
1102                 extent = btrfs_item_ptr(leaf, path->slots[0],
1103                                         struct btrfs_file_extent_item);
1104
1105                 type = btrfs_file_extent_type(leaf, extent);
1106                 if (type == BTRFS_FILE_EXTENT_REG &&
1107                     btrfs_file_extent_num_bytes(leaf, extent) < thresh &&
1108                     check_defrag_in_cache(inode, min_key.offset, thresh)) {
1109                         *off = min_key.offset;
1110                         btrfs_free_path(path);
1111                         return 0;
1112                 }
1113
1114                 path->slots[0]++;
1115                 if (path->slots[0] < btrfs_header_nritems(leaf)) {
1116                         btrfs_item_key_to_cpu(leaf, &min_key, path->slots[0]);
1117                         goto process_slot;
1118                 }
1119
1120                 if (min_key.offset == (u64)-1)
1121                         goto none;
1122
1123                 min_key.offset++;
1124                 btrfs_release_path(path);
1125         }
1126 none:
1127         btrfs_free_path(path);
1128         return -ENOENT;
1129 }
1130
1131 static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start)
1132 {
1133         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1134         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1135         struct extent_map *em;
1136         u64 len = PAGE_SIZE;
1137
1138         /*
1139          * hopefully we have this extent in the tree already, try without
1140          * the full extent lock
1141          */
1142         read_lock(&em_tree->lock);
1143         em = lookup_extent_mapping(em_tree, start, len);
1144         read_unlock(&em_tree->lock);
1145
1146         if (!em) {
1147                 struct extent_state *cached = NULL;
1148                 u64 end = start + len - 1;
1149
1150                 /* get the big lock and read metadata off disk */
1151                 lock_extent_bits(io_tree, start, end, &cached);
1152                 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
1153                 unlock_extent_cached(io_tree, start, end, &cached);
1154
1155                 if (IS_ERR(em))
1156                         return NULL;
1157         }
1158
1159         return em;
1160 }
1161
1162 static bool defrag_check_next_extent(struct inode *inode, struct extent_map *em)
1163 {
1164         struct extent_map *next;
1165         bool ret = true;
1166
1167         /* this is the last extent */
1168         if (em->start + em->len >= i_size_read(inode))
1169                 return false;
1170
1171         next = defrag_lookup_extent(inode, em->start + em->len);
1172         if (!next || next->block_start >= EXTENT_MAP_LAST_BYTE)
1173                 ret = false;
1174         else if ((em->block_start + em->block_len == next->block_start) &&
1175                  (em->block_len > SZ_128K && next->block_len > SZ_128K))
1176                 ret = false;
1177
1178         free_extent_map(next);
1179         return ret;
1180 }
1181
1182 static int should_defrag_range(struct inode *inode, u64 start, u32 thresh,
1183                                u64 *last_len, u64 *skip, u64 *defrag_end,
1184                                int compress)
1185 {
1186         struct extent_map *em;
1187         int ret = 1;
1188         bool next_mergeable = true;
1189         bool prev_mergeable = true;
1190
1191         /*
1192          * make sure that once we start defragging an extent, we keep on
1193          * defragging it
1194          */
1195         if (start < *defrag_end)
1196                 return 1;
1197
1198         *skip = 0;
1199
1200         em = defrag_lookup_extent(inode, start);
1201         if (!em)
1202                 return 0;
1203
1204         /* this will cover holes, and inline extents */
1205         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1206                 ret = 0;
1207                 goto out;
1208         }
1209
1210         if (!*defrag_end)
1211                 prev_mergeable = false;
1212
1213         next_mergeable = defrag_check_next_extent(inode, em);
1214         /*
1215          * we hit a real extent, if it is big or the next extent is not a
1216          * real extent, don't bother defragging it
1217          */
1218         if (!compress && (*last_len == 0 || *last_len >= thresh) &&
1219             (em->len >= thresh || (!next_mergeable && !prev_mergeable)))
1220                 ret = 0;
1221 out:
1222         /*
1223          * last_len ends up being a counter of how many bytes we've defragged.
1224          * every time we choose not to defrag an extent, we reset *last_len
1225          * so that the next tiny extent will force a defrag.
1226          *
1227          * The end result of this is that tiny extents before a single big
1228          * extent will force at least part of that big extent to be defragged.
1229          */
1230         if (ret) {
1231                 *defrag_end = extent_map_end(em);
1232         } else {
1233                 *last_len = 0;
1234                 *skip = extent_map_end(em);
1235                 *defrag_end = 0;
1236         }
1237
1238         free_extent_map(em);
1239         return ret;
1240 }
1241
1242 /*
1243  * it doesn't do much good to defrag one or two pages
1244  * at a time.  This pulls in a nice chunk of pages
1245  * to COW and defrag.
1246  *
1247  * It also makes sure the delalloc code has enough
1248  * dirty data to avoid making new small extents as part
1249  * of the defrag
1250  *
1251  * It's a good idea to start RA on this range
1252  * before calling this.
1253  */
1254 static int cluster_pages_for_defrag(struct inode *inode,
1255                                     struct page **pages,
1256                                     unsigned long start_index,
1257                                     unsigned long num_pages)
1258 {
1259         unsigned long file_end;
1260         u64 isize = i_size_read(inode);
1261         u64 page_start;
1262         u64 page_end;
1263         u64 page_cnt;
1264         int ret;
1265         int i;
1266         int i_done;
1267         struct btrfs_ordered_extent *ordered;
1268         struct extent_state *cached_state = NULL;
1269         struct extent_io_tree *tree;
1270         struct extent_changeset *data_reserved = NULL;
1271         gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1272
1273         file_end = (isize - 1) >> PAGE_SHIFT;
1274         if (!isize || start_index > file_end)
1275                 return 0;
1276
1277         page_cnt = min_t(u64, (u64)num_pages, (u64)file_end - start_index + 1);
1278
1279         ret = btrfs_delalloc_reserve_space(BTRFS_I(inode), &data_reserved,
1280                         start_index << PAGE_SHIFT,
1281                         page_cnt << PAGE_SHIFT);
1282         if (ret)
1283                 return ret;
1284         i_done = 0;
1285         tree = &BTRFS_I(inode)->io_tree;
1286
1287         /* step one, lock all the pages */
1288         for (i = 0; i < page_cnt; i++) {
1289                 struct page *page;
1290 again:
1291                 page = find_or_create_page(inode->i_mapping,
1292                                            start_index + i, mask);
1293                 if (!page)
1294                         break;
1295
1296                 page_start = page_offset(page);
1297                 page_end = page_start + PAGE_SIZE - 1;
1298                 while (1) {
1299                         lock_extent_bits(tree, page_start, page_end,
1300                                          &cached_state);
1301                         ordered = btrfs_lookup_ordered_extent(BTRFS_I(inode),
1302                                                               page_start);
1303                         unlock_extent_cached(tree, page_start, page_end,
1304                                              &cached_state);
1305                         if (!ordered)
1306                                 break;
1307
1308                         unlock_page(page);
1309                         btrfs_start_ordered_extent(inode, ordered, 1);
1310                         btrfs_put_ordered_extent(ordered);
1311                         lock_page(page);
1312                         /*
1313                          * we unlocked the page above, so we need check if
1314                          * it was released or not.
1315                          */
1316                         if (page->mapping != inode->i_mapping) {
1317                                 unlock_page(page);
1318                                 put_page(page);
1319                                 goto again;
1320                         }
1321                 }
1322
1323                 if (!PageUptodate(page)) {
1324                         btrfs_readpage(NULL, page);
1325                         lock_page(page);
1326                         if (!PageUptodate(page)) {
1327                                 unlock_page(page);
1328                                 put_page(page);
1329                                 ret = -EIO;
1330                                 break;
1331                         }
1332                 }
1333
1334                 if (page->mapping != inode->i_mapping) {
1335                         unlock_page(page);
1336                         put_page(page);
1337                         goto again;
1338                 }
1339
1340                 pages[i] = page;
1341                 i_done++;
1342         }
1343         if (!i_done || ret)
1344                 goto out;
1345
1346         if (!(inode->i_sb->s_flags & SB_ACTIVE))
1347                 goto out;
1348
1349         /*
1350          * so now we have a nice long stream of locked
1351          * and up to date pages, lets wait on them
1352          */
1353         for (i = 0; i < i_done; i++)
1354                 wait_on_page_writeback(pages[i]);
1355
1356         page_start = page_offset(pages[0]);
1357         page_end = page_offset(pages[i_done - 1]) + PAGE_SIZE;
1358
1359         lock_extent_bits(&BTRFS_I(inode)->io_tree,
1360                          page_start, page_end - 1, &cached_state);
1361         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start,
1362                           page_end - 1, EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
1363                           EXTENT_DEFRAG, 0, 0, &cached_state);
1364
1365         if (i_done != page_cnt) {
1366                 spin_lock(&BTRFS_I(inode)->lock);
1367                 btrfs_mod_outstanding_extents(BTRFS_I(inode), 1);
1368                 spin_unlock(&BTRFS_I(inode)->lock);
1369                 btrfs_delalloc_release_space(BTRFS_I(inode), data_reserved,
1370                                 start_index << PAGE_SHIFT,
1371                                 (page_cnt - i_done) << PAGE_SHIFT, true);
1372         }
1373
1374
1375         set_extent_defrag(&BTRFS_I(inode)->io_tree, page_start, page_end - 1,
1376                           &cached_state);
1377
1378         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1379                              page_start, page_end - 1, &cached_state);
1380
1381         for (i = 0; i < i_done; i++) {
1382                 clear_page_dirty_for_io(pages[i]);
1383                 ClearPageChecked(pages[i]);
1384                 set_page_extent_mapped(pages[i]);
1385                 set_page_dirty(pages[i]);
1386                 unlock_page(pages[i]);
1387                 put_page(pages[i]);
1388         }
1389         btrfs_delalloc_release_extents(BTRFS_I(inode), page_cnt << PAGE_SHIFT);
1390         extent_changeset_free(data_reserved);
1391         return i_done;
1392 out:
1393         for (i = 0; i < i_done; i++) {
1394                 unlock_page(pages[i]);
1395                 put_page(pages[i]);
1396         }
1397         btrfs_delalloc_release_space(BTRFS_I(inode), data_reserved,
1398                         start_index << PAGE_SHIFT,
1399                         page_cnt << PAGE_SHIFT, true);
1400         btrfs_delalloc_release_extents(BTRFS_I(inode), page_cnt << PAGE_SHIFT);
1401         extent_changeset_free(data_reserved);
1402         return ret;
1403
1404 }
1405
1406 int btrfs_defrag_file(struct inode *inode, struct file *file,
1407                       struct btrfs_ioctl_defrag_range_args *range,
1408                       u64 newer_than, unsigned long max_to_defrag)
1409 {
1410         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1411         struct btrfs_root *root = BTRFS_I(inode)->root;
1412         struct file_ra_state *ra = NULL;
1413         unsigned long last_index;
1414         u64 isize = i_size_read(inode);
1415         u64 last_len = 0;
1416         u64 skip = 0;
1417         u64 defrag_end = 0;
1418         u64 newer_off = range->start;
1419         unsigned long i;
1420         unsigned long ra_index = 0;
1421         int ret;
1422         int defrag_count = 0;
1423         int compress_type = BTRFS_COMPRESS_ZLIB;
1424         u32 extent_thresh = range->extent_thresh;
1425         unsigned long max_cluster = SZ_256K >> PAGE_SHIFT;
1426         unsigned long cluster = max_cluster;
1427         u64 new_align = ~((u64)SZ_128K - 1);
1428         struct page **pages = NULL;
1429         bool do_compress = range->flags & BTRFS_DEFRAG_RANGE_COMPRESS;
1430
1431         if (isize == 0)
1432                 return 0;
1433
1434         if (range->start >= isize)
1435                 return -EINVAL;
1436
1437         if (do_compress) {
1438                 if (range->compress_type >= BTRFS_NR_COMPRESS_TYPES)
1439                         return -EINVAL;
1440                 if (range->compress_type)
1441                         compress_type = range->compress_type;
1442         }
1443
1444         if (extent_thresh == 0)
1445                 extent_thresh = SZ_256K;
1446
1447         /*
1448          * If we were not given a file, allocate a readahead context. As
1449          * readahead is just an optimization, defrag will work without it so
1450          * we don't error out.
1451          */
1452         if (!file) {
1453                 ra = kzalloc(sizeof(*ra), GFP_KERNEL);
1454                 if (ra)
1455                         file_ra_state_init(ra, inode->i_mapping);
1456         } else {
1457                 ra = &file->f_ra;
1458         }
1459
1460         pages = kmalloc_array(max_cluster, sizeof(struct page *), GFP_KERNEL);
1461         if (!pages) {
1462                 ret = -ENOMEM;
1463                 goto out_ra;
1464         }
1465
1466         /* find the last page to defrag */
1467         if (range->start + range->len > range->start) {
1468                 last_index = min_t(u64, isize - 1,
1469                          range->start + range->len - 1) >> PAGE_SHIFT;
1470         } else {
1471                 last_index = (isize - 1) >> PAGE_SHIFT;
1472         }
1473
1474         if (newer_than) {
1475                 ret = find_new_extents(root, inode, newer_than,
1476                                        &newer_off, SZ_64K);
1477                 if (!ret) {
1478                         range->start = newer_off;
1479                         /*
1480                          * we always align our defrag to help keep
1481                          * the extents in the file evenly spaced
1482                          */
1483                         i = (newer_off & new_align) >> PAGE_SHIFT;
1484                 } else
1485                         goto out_ra;
1486         } else {
1487                 i = range->start >> PAGE_SHIFT;
1488         }
1489         if (!max_to_defrag)
1490                 max_to_defrag = last_index - i + 1;
1491
1492         /*
1493          * make writeback starts from i, so the defrag range can be
1494          * written sequentially.
1495          */
1496         if (i < inode->i_mapping->writeback_index)
1497                 inode->i_mapping->writeback_index = i;
1498
1499         while (i <= last_index && defrag_count < max_to_defrag &&
1500                (i < DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE))) {
1501                 /*
1502                  * make sure we stop running if someone unmounts
1503                  * the FS
1504                  */
1505                 if (!(inode->i_sb->s_flags & SB_ACTIVE))
1506                         break;
1507
1508                 if (btrfs_defrag_cancelled(fs_info)) {
1509                         btrfs_debug(fs_info, "defrag_file cancelled");
1510                         ret = -EAGAIN;
1511                         break;
1512                 }
1513
1514                 if (!should_defrag_range(inode, (u64)i << PAGE_SHIFT,
1515                                          extent_thresh, &last_len, &skip,
1516                                          &defrag_end, do_compress)){
1517                         unsigned long next;
1518                         /*
1519                          * the should_defrag function tells us how much to skip
1520                          * bump our counter by the suggested amount
1521                          */
1522                         next = DIV_ROUND_UP(skip, PAGE_SIZE);
1523                         i = max(i + 1, next);
1524                         continue;
1525                 }
1526
1527                 if (!newer_than) {
1528                         cluster = (PAGE_ALIGN(defrag_end) >>
1529                                    PAGE_SHIFT) - i;
1530                         cluster = min(cluster, max_cluster);
1531                 } else {
1532                         cluster = max_cluster;
1533                 }
1534
1535                 if (i + cluster > ra_index) {
1536                         ra_index = max(i, ra_index);
1537                         if (ra)
1538                                 page_cache_sync_readahead(inode->i_mapping, ra,
1539                                                 file, ra_index, cluster);
1540                         ra_index += cluster;
1541                 }
1542
1543                 inode_lock(inode);
1544                 if (IS_SWAPFILE(inode)) {
1545                         ret = -ETXTBSY;
1546                 } else {
1547                         if (do_compress)
1548                                 BTRFS_I(inode)->defrag_compress = compress_type;
1549                         ret = cluster_pages_for_defrag(inode, pages, i, cluster);
1550                 }
1551                 if (ret < 0) {
1552                         inode_unlock(inode);
1553                         goto out_ra;
1554                 }
1555
1556                 defrag_count += ret;
1557                 balance_dirty_pages_ratelimited(inode->i_mapping);
1558                 inode_unlock(inode);
1559
1560                 if (newer_than) {
1561                         if (newer_off == (u64)-1)
1562                                 break;
1563
1564                         if (ret > 0)
1565                                 i += ret;
1566
1567                         newer_off = max(newer_off + 1,
1568                                         (u64)i << PAGE_SHIFT);
1569
1570                         ret = find_new_extents(root, inode, newer_than,
1571                                                &newer_off, SZ_64K);
1572                         if (!ret) {
1573                                 range->start = newer_off;
1574                                 i = (newer_off & new_align) >> PAGE_SHIFT;
1575                         } else {
1576                                 break;
1577                         }
1578                 } else {
1579                         if (ret > 0) {
1580                                 i += ret;
1581                                 last_len += ret << PAGE_SHIFT;
1582                         } else {
1583                                 i++;
1584                                 last_len = 0;
1585                         }
1586                 }
1587         }
1588
1589         if ((range->flags & BTRFS_DEFRAG_RANGE_START_IO)) {
1590                 filemap_flush(inode->i_mapping);
1591                 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1592                              &BTRFS_I(inode)->runtime_flags))
1593                         filemap_flush(inode->i_mapping);
1594         }
1595
1596         if (range->compress_type == BTRFS_COMPRESS_LZO) {
1597                 btrfs_set_fs_incompat(fs_info, COMPRESS_LZO);
1598         } else if (range->compress_type == BTRFS_COMPRESS_ZSTD) {
1599                 btrfs_set_fs_incompat(fs_info, COMPRESS_ZSTD);
1600         }
1601
1602         ret = defrag_count;
1603
1604 out_ra:
1605         if (do_compress) {
1606                 inode_lock(inode);
1607                 BTRFS_I(inode)->defrag_compress = BTRFS_COMPRESS_NONE;
1608                 inode_unlock(inode);
1609         }
1610         if (!file)
1611                 kfree(ra);
1612         kfree(pages);
1613         return ret;
1614 }
1615
1616 static noinline int btrfs_ioctl_resize(struct file *file,
1617                                         void __user *arg)
1618 {
1619         struct inode *inode = file_inode(file);
1620         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1621         u64 new_size;
1622         u64 old_size;
1623         u64 devid = 1;
1624         struct btrfs_root *root = BTRFS_I(inode)->root;
1625         struct btrfs_ioctl_vol_args *vol_args;
1626         struct btrfs_trans_handle *trans;
1627         struct btrfs_device *device = NULL;
1628         char *sizestr;
1629         char *retptr;
1630         char *devstr = NULL;
1631         int ret = 0;
1632         int mod = 0;
1633
1634         if (!capable(CAP_SYS_ADMIN))
1635                 return -EPERM;
1636
1637         ret = mnt_want_write_file(file);
1638         if (ret)
1639                 return ret;
1640
1641         if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
1642                 mnt_drop_write_file(file);
1643                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
1644         }
1645
1646         vol_args = memdup_user(arg, sizeof(*vol_args));
1647         if (IS_ERR(vol_args)) {
1648                 ret = PTR_ERR(vol_args);
1649                 goto out;
1650         }
1651
1652         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1653
1654         sizestr = vol_args->name;
1655         devstr = strchr(sizestr, ':');
1656         if (devstr) {
1657                 sizestr = devstr + 1;
1658                 *devstr = '\0';
1659                 devstr = vol_args->name;
1660                 ret = kstrtoull(devstr, 10, &devid);
1661                 if (ret)
1662                         goto out_free;
1663                 if (!devid) {
1664                         ret = -EINVAL;
1665                         goto out_free;
1666                 }
1667                 btrfs_info(fs_info, "resizing devid %llu", devid);
1668         }
1669
1670         device = btrfs_find_device(fs_info->fs_devices, devid, NULL, NULL, true);
1671         if (!device) {
1672                 btrfs_info(fs_info, "resizer unable to find device %llu",
1673                            devid);
1674                 ret = -ENODEV;
1675                 goto out_free;
1676         }
1677
1678         if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1679                 btrfs_info(fs_info,
1680                            "resizer unable to apply on readonly device %llu",
1681                        devid);
1682                 ret = -EPERM;
1683                 goto out_free;
1684         }
1685
1686         if (!strcmp(sizestr, "max"))
1687                 new_size = device->bdev->bd_inode->i_size;
1688         else {
1689                 if (sizestr[0] == '-') {
1690                         mod = -1;
1691                         sizestr++;
1692                 } else if (sizestr[0] == '+') {
1693                         mod = 1;
1694                         sizestr++;
1695                 }
1696                 new_size = memparse(sizestr, &retptr);
1697                 if (*retptr != '\0' || new_size == 0) {
1698                         ret = -EINVAL;
1699                         goto out_free;
1700                 }
1701         }
1702
1703         if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1704                 ret = -EPERM;
1705                 goto out_free;
1706         }
1707
1708         old_size = btrfs_device_get_total_bytes(device);
1709
1710         if (mod < 0) {
1711                 if (new_size > old_size) {
1712                         ret = -EINVAL;
1713                         goto out_free;
1714                 }
1715                 new_size = old_size - new_size;
1716         } else if (mod > 0) {
1717                 if (new_size > ULLONG_MAX - old_size) {
1718                         ret = -ERANGE;
1719                         goto out_free;
1720                 }
1721                 new_size = old_size + new_size;
1722         }
1723
1724         if (new_size < SZ_256M) {
1725                 ret = -EINVAL;
1726                 goto out_free;
1727         }
1728         if (new_size > device->bdev->bd_inode->i_size) {
1729                 ret = -EFBIG;
1730                 goto out_free;
1731         }
1732
1733         new_size = round_down(new_size, fs_info->sectorsize);
1734
1735         if (new_size > old_size) {
1736                 trans = btrfs_start_transaction(root, 0);
1737                 if (IS_ERR(trans)) {
1738                         ret = PTR_ERR(trans);
1739                         goto out_free;
1740                 }
1741                 ret = btrfs_grow_device(trans, device, new_size);
1742                 btrfs_commit_transaction(trans);
1743         } else if (new_size < old_size) {
1744                 ret = btrfs_shrink_device(device, new_size);
1745         } /* equal, nothing need to do */
1746
1747         if (ret == 0 && new_size != old_size)
1748                 btrfs_info_in_rcu(fs_info,
1749                         "resize device %s (devid %llu) from %llu to %llu",
1750                         rcu_str_deref(device->name), device->devid,
1751                         old_size, new_size);
1752 out_free:
1753         kfree(vol_args);
1754 out:
1755         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
1756         mnt_drop_write_file(file);
1757         return ret;
1758 }
1759
1760 static noinline int __btrfs_ioctl_snap_create(struct file *file,
1761                                 const char *name, unsigned long fd, int subvol,
1762                                 bool readonly,
1763                                 struct btrfs_qgroup_inherit *inherit)
1764 {
1765         int namelen;
1766         int ret = 0;
1767
1768         if (!S_ISDIR(file_inode(file)->i_mode))
1769                 return -ENOTDIR;
1770
1771         ret = mnt_want_write_file(file);
1772         if (ret)
1773                 goto out;
1774
1775         namelen = strlen(name);
1776         if (strchr(name, '/')) {
1777                 ret = -EINVAL;
1778                 goto out_drop_write;
1779         }
1780
1781         if (name[0] == '.' &&
1782            (namelen == 1 || (name[1] == '.' && namelen == 2))) {
1783                 ret = -EEXIST;
1784                 goto out_drop_write;
1785         }
1786
1787         if (subvol) {
1788                 ret = btrfs_mksubvol(&file->f_path, name, namelen,
1789                                      NULL, readonly, inherit);
1790         } else {
1791                 struct fd src = fdget(fd);
1792                 struct inode *src_inode;
1793                 if (!src.file) {
1794                         ret = -EINVAL;
1795                         goto out_drop_write;
1796                 }
1797
1798                 src_inode = file_inode(src.file);
1799                 if (src_inode->i_sb != file_inode(file)->i_sb) {
1800                         btrfs_info(BTRFS_I(file_inode(file))->root->fs_info,
1801                                    "Snapshot src from another FS");
1802                         ret = -EXDEV;
1803                 } else if (!inode_owner_or_capable(src_inode)) {
1804                         /*
1805                          * Subvolume creation is not restricted, but snapshots
1806                          * are limited to own subvolumes only
1807                          */
1808                         ret = -EPERM;
1809                 } else {
1810                         ret = btrfs_mksnapshot(&file->f_path, name, namelen,
1811                                              BTRFS_I(src_inode)->root,
1812                                              readonly, inherit);
1813                 }
1814                 fdput(src);
1815         }
1816 out_drop_write:
1817         mnt_drop_write_file(file);
1818 out:
1819         return ret;
1820 }
1821
1822 static noinline int btrfs_ioctl_snap_create(struct file *file,
1823                                             void __user *arg, int subvol)
1824 {
1825         struct btrfs_ioctl_vol_args *vol_args;
1826         int ret;
1827
1828         if (!S_ISDIR(file_inode(file)->i_mode))
1829                 return -ENOTDIR;
1830
1831         vol_args = memdup_user(arg, sizeof(*vol_args));
1832         if (IS_ERR(vol_args))
1833                 return PTR_ERR(vol_args);
1834         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
1835
1836         ret = __btrfs_ioctl_snap_create(file, vol_args->name, vol_args->fd,
1837                                         subvol, false, NULL);
1838
1839         kfree(vol_args);
1840         return ret;
1841 }
1842
1843 static noinline int btrfs_ioctl_snap_create_v2(struct file *file,
1844                                                void __user *arg, int subvol)
1845 {
1846         struct btrfs_ioctl_vol_args_v2 *vol_args;
1847         int ret;
1848         bool readonly = false;
1849         struct btrfs_qgroup_inherit *inherit = NULL;
1850
1851         if (!S_ISDIR(file_inode(file)->i_mode))
1852                 return -ENOTDIR;
1853
1854         vol_args = memdup_user(arg, sizeof(*vol_args));
1855         if (IS_ERR(vol_args))
1856                 return PTR_ERR(vol_args);
1857         vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
1858
1859         if (vol_args->flags & ~BTRFS_SUBVOL_CREATE_ARGS_MASK) {
1860                 ret = -EOPNOTSUPP;
1861                 goto free_args;
1862         }
1863
1864         if (vol_args->flags & BTRFS_SUBVOL_RDONLY)
1865                 readonly = true;
1866         if (vol_args->flags & BTRFS_SUBVOL_QGROUP_INHERIT) {
1867                 if (vol_args->size > PAGE_SIZE) {
1868                         ret = -EINVAL;
1869                         goto free_args;
1870                 }
1871                 inherit = memdup_user(vol_args->qgroup_inherit, vol_args->size);
1872                 if (IS_ERR(inherit)) {
1873                         ret = PTR_ERR(inherit);
1874                         goto free_args;
1875                 }
1876         }
1877
1878         ret = __btrfs_ioctl_snap_create(file, vol_args->name, vol_args->fd,
1879                                         subvol, readonly, inherit);
1880         if (ret)
1881                 goto free_inherit;
1882 free_inherit:
1883         kfree(inherit);
1884 free_args:
1885         kfree(vol_args);
1886         return ret;
1887 }
1888
1889 static noinline int btrfs_ioctl_subvol_getflags(struct file *file,
1890                                                 void __user *arg)
1891 {
1892         struct inode *inode = file_inode(file);
1893         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1894         struct btrfs_root *root = BTRFS_I(inode)->root;
1895         int ret = 0;
1896         u64 flags = 0;
1897
1898         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID)
1899                 return -EINVAL;
1900
1901         down_read(&fs_info->subvol_sem);
1902         if (btrfs_root_readonly(root))
1903                 flags |= BTRFS_SUBVOL_RDONLY;
1904         up_read(&fs_info->subvol_sem);
1905
1906         if (copy_to_user(arg, &flags, sizeof(flags)))
1907                 ret = -EFAULT;
1908
1909         return ret;
1910 }
1911
1912 static noinline int btrfs_ioctl_subvol_setflags(struct file *file,
1913                                               void __user *arg)
1914 {
1915         struct inode *inode = file_inode(file);
1916         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1917         struct btrfs_root *root = BTRFS_I(inode)->root;
1918         struct btrfs_trans_handle *trans;
1919         u64 root_flags;
1920         u64 flags;
1921         int ret = 0;
1922
1923         if (!inode_owner_or_capable(inode))
1924                 return -EPERM;
1925
1926         ret = mnt_want_write_file(file);
1927         if (ret)
1928                 goto out;
1929
1930         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
1931                 ret = -EINVAL;
1932                 goto out_drop_write;
1933         }
1934
1935         if (copy_from_user(&flags, arg, sizeof(flags))) {
1936                 ret = -EFAULT;
1937                 goto out_drop_write;
1938         }
1939
1940         if (flags & ~BTRFS_SUBVOL_RDONLY) {
1941                 ret = -EOPNOTSUPP;
1942                 goto out_drop_write;
1943         }
1944
1945         down_write(&fs_info->subvol_sem);
1946
1947         /* nothing to do */
1948         if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root))
1949                 goto out_drop_sem;
1950
1951         root_flags = btrfs_root_flags(&root->root_item);
1952         if (flags & BTRFS_SUBVOL_RDONLY) {
1953                 btrfs_set_root_flags(&root->root_item,
1954                                      root_flags | BTRFS_ROOT_SUBVOL_RDONLY);
1955         } else {
1956                 /*
1957                  * Block RO -> RW transition if this subvolume is involved in
1958                  * send
1959                  */
1960                 spin_lock(&root->root_item_lock);
1961                 if (root->send_in_progress == 0) {
1962                         btrfs_set_root_flags(&root->root_item,
1963                                      root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY);
1964                         spin_unlock(&root->root_item_lock);
1965                 } else {
1966                         spin_unlock(&root->root_item_lock);
1967                         btrfs_warn(fs_info,
1968                                    "Attempt to set subvolume %llu read-write during send",
1969                                    root->root_key.objectid);
1970                         ret = -EPERM;
1971                         goto out_drop_sem;
1972                 }
1973         }
1974
1975         trans = btrfs_start_transaction(root, 1);
1976         if (IS_ERR(trans)) {
1977                 ret = PTR_ERR(trans);
1978                 goto out_reset;
1979         }
1980
1981         ret = btrfs_update_root(trans, fs_info->tree_root,
1982                                 &root->root_key, &root->root_item);
1983         if (ret < 0) {
1984                 btrfs_end_transaction(trans);
1985                 goto out_reset;
1986         }
1987
1988         ret = btrfs_commit_transaction(trans);
1989
1990 out_reset:
1991         if (ret)
1992                 btrfs_set_root_flags(&root->root_item, root_flags);
1993 out_drop_sem:
1994         up_write(&fs_info->subvol_sem);
1995 out_drop_write:
1996         mnt_drop_write_file(file);
1997 out:
1998         return ret;
1999 }
2000
2001 static noinline int key_in_sk(struct btrfs_key *key,
2002                               struct btrfs_ioctl_search_key *sk)
2003 {
2004         struct btrfs_key test;
2005         int ret;
2006
2007         test.objectid = sk->min_objectid;
2008         test.type = sk->min_type;
2009         test.offset = sk->min_offset;
2010
2011         ret = btrfs_comp_cpu_keys(key, &test);
2012         if (ret < 0)
2013                 return 0;
2014
2015         test.objectid = sk->max_objectid;
2016         test.type = sk->max_type;
2017         test.offset = sk->max_offset;
2018
2019         ret = btrfs_comp_cpu_keys(key, &test);
2020         if (ret > 0)
2021                 return 0;
2022         return 1;
2023 }
2024
2025 static noinline int copy_to_sk(struct btrfs_path *path,
2026                                struct btrfs_key *key,
2027                                struct btrfs_ioctl_search_key *sk,
2028                                size_t *buf_size,
2029                                char __user *ubuf,
2030                                unsigned long *sk_offset,
2031                                int *num_found)
2032 {
2033         u64 found_transid;
2034         struct extent_buffer *leaf;
2035         struct btrfs_ioctl_search_header sh;
2036         struct btrfs_key test;
2037         unsigned long item_off;
2038         unsigned long item_len;
2039         int nritems;
2040         int i;
2041         int slot;
2042         int ret = 0;
2043
2044         leaf = path->nodes[0];
2045         slot = path->slots[0];
2046         nritems = btrfs_header_nritems(leaf);
2047
2048         if (btrfs_header_generation(leaf) > sk->max_transid) {
2049                 i = nritems;
2050                 goto advance_key;
2051         }
2052         found_transid = btrfs_header_generation(leaf);
2053
2054         for (i = slot; i < nritems; i++) {
2055                 item_off = btrfs_item_ptr_offset(leaf, i);
2056                 item_len = btrfs_item_size_nr(leaf, i);
2057
2058                 btrfs_item_key_to_cpu(leaf, key, i);
2059                 if (!key_in_sk(key, sk))
2060                         continue;
2061
2062                 if (sizeof(sh) + item_len > *buf_size) {
2063                         if (*num_found) {
2064                                 ret = 1;
2065                                 goto out;
2066                         }
2067
2068                         /*
2069                          * return one empty item back for v1, which does not
2070                          * handle -EOVERFLOW
2071                          */
2072
2073                         *buf_size = sizeof(sh) + item_len;
2074                         item_len = 0;
2075                         ret = -EOVERFLOW;
2076                 }
2077
2078                 if (sizeof(sh) + item_len + *sk_offset > *buf_size) {
2079                         ret = 1;
2080                         goto out;
2081                 }
2082
2083                 sh.objectid = key->objectid;
2084                 sh.offset = key->offset;
2085                 sh.type = key->type;
2086                 sh.len = item_len;
2087                 sh.transid = found_transid;
2088
2089                 /* copy search result header */
2090                 if (copy_to_user(ubuf + *sk_offset, &sh, sizeof(sh))) {
2091                         ret = -EFAULT;
2092                         goto out;
2093                 }
2094
2095                 *sk_offset += sizeof(sh);
2096
2097                 if (item_len) {
2098                         char __user *up = ubuf + *sk_offset;
2099                         /* copy the item */
2100                         if (read_extent_buffer_to_user(leaf, up,
2101                                                        item_off, item_len)) {
2102                                 ret = -EFAULT;
2103                                 goto out;
2104                         }
2105
2106                         *sk_offset += item_len;
2107                 }
2108                 (*num_found)++;
2109
2110                 if (ret) /* -EOVERFLOW from above */
2111                         goto out;
2112
2113                 if (*num_found >= sk->nr_items) {
2114                         ret = 1;
2115                         goto out;
2116                 }
2117         }
2118 advance_key:
2119         ret = 0;
2120         test.objectid = sk->max_objectid;
2121         test.type = sk->max_type;
2122         test.offset = sk->max_offset;
2123         if (btrfs_comp_cpu_keys(key, &test) >= 0)
2124                 ret = 1;
2125         else if (key->offset < (u64)-1)
2126                 key->offset++;
2127         else if (key->type < (u8)-1) {
2128                 key->offset = 0;
2129                 key->type++;
2130         } else if (key->objectid < (u64)-1) {
2131                 key->offset = 0;
2132                 key->type = 0;
2133                 key->objectid++;
2134         } else
2135                 ret = 1;
2136 out:
2137         /*
2138          *  0: all items from this leaf copied, continue with next
2139          *  1: * more items can be copied, but unused buffer is too small
2140          *     * all items were found
2141          *     Either way, it will stops the loop which iterates to the next
2142          *     leaf
2143          *  -EOVERFLOW: item was to large for buffer
2144          *  -EFAULT: could not copy extent buffer back to userspace
2145          */
2146         return ret;
2147 }
2148
2149 static noinline int search_ioctl(struct inode *inode,
2150                                  struct btrfs_ioctl_search_key *sk,
2151                                  size_t *buf_size,
2152                                  char __user *ubuf)
2153 {
2154         struct btrfs_fs_info *info = btrfs_sb(inode->i_sb);
2155         struct btrfs_root *root;
2156         struct btrfs_key key;
2157         struct btrfs_path *path;
2158         int ret;
2159         int num_found = 0;
2160         unsigned long sk_offset = 0;
2161
2162         if (*buf_size < sizeof(struct btrfs_ioctl_search_header)) {
2163                 *buf_size = sizeof(struct btrfs_ioctl_search_header);
2164                 return -EOVERFLOW;
2165         }
2166
2167         path = btrfs_alloc_path();
2168         if (!path)
2169                 return -ENOMEM;
2170
2171         if (sk->tree_id == 0) {
2172                 /* search the root of the inode that was passed */
2173                 root = btrfs_grab_root(BTRFS_I(inode)->root);
2174         } else {
2175                 root = btrfs_get_fs_root(info, sk->tree_id, true);
2176                 if (IS_ERR(root)) {
2177                         btrfs_free_path(path);
2178                         return PTR_ERR(root);
2179                 }
2180         }
2181
2182         key.objectid = sk->min_objectid;
2183         key.type = sk->min_type;
2184         key.offset = sk->min_offset;
2185
2186         while (1) {
2187                 ret = btrfs_search_forward(root, &key, path, sk->min_transid);
2188                 if (ret != 0) {
2189                         if (ret > 0)
2190                                 ret = 0;
2191                         goto err;
2192                 }
2193                 ret = copy_to_sk(path, &key, sk, buf_size, ubuf,
2194                                  &sk_offset, &num_found);
2195                 btrfs_release_path(path);
2196                 if (ret)
2197                         break;
2198
2199         }
2200         if (ret > 0)
2201                 ret = 0;
2202 err:
2203         sk->nr_items = num_found;
2204         btrfs_put_root(root);
2205         btrfs_free_path(path);
2206         return ret;
2207 }
2208
2209 static noinline int btrfs_ioctl_tree_search(struct file *file,
2210                                            void __user *argp)
2211 {
2212         struct btrfs_ioctl_search_args __user *uargs;
2213         struct btrfs_ioctl_search_key sk;
2214         struct inode *inode;
2215         int ret;
2216         size_t buf_size;
2217
2218         if (!capable(CAP_SYS_ADMIN))
2219                 return -EPERM;
2220
2221         uargs = (struct btrfs_ioctl_search_args __user *)argp;
2222
2223         if (copy_from_user(&sk, &uargs->key, sizeof(sk)))
2224                 return -EFAULT;
2225
2226         buf_size = sizeof(uargs->buf);
2227
2228         inode = file_inode(file);
2229         ret = search_ioctl(inode, &sk, &buf_size, uargs->buf);
2230
2231         /*
2232          * In the origin implementation an overflow is handled by returning a
2233          * search header with a len of zero, so reset ret.
2234          */
2235         if (ret == -EOVERFLOW)
2236                 ret = 0;
2237
2238         if (ret == 0 && copy_to_user(&uargs->key, &sk, sizeof(sk)))
2239                 ret = -EFAULT;
2240         return ret;
2241 }
2242
2243 static noinline int btrfs_ioctl_tree_search_v2(struct file *file,
2244                                                void __user *argp)
2245 {
2246         struct btrfs_ioctl_search_args_v2 __user *uarg;
2247         struct btrfs_ioctl_search_args_v2 args;
2248         struct inode *inode;
2249         int ret;
2250         size_t buf_size;
2251         const size_t buf_limit = SZ_16M;
2252
2253         if (!capable(CAP_SYS_ADMIN))
2254                 return -EPERM;
2255
2256         /* copy search header and buffer size */
2257         uarg = (struct btrfs_ioctl_search_args_v2 __user *)argp;
2258         if (copy_from_user(&args, uarg, sizeof(args)))
2259                 return -EFAULT;
2260
2261         buf_size = args.buf_size;
2262
2263         /* limit result size to 16MB */
2264         if (buf_size > buf_limit)
2265                 buf_size = buf_limit;
2266
2267         inode = file_inode(file);
2268         ret = search_ioctl(inode, &args.key, &buf_size,
2269                            (char __user *)(&uarg->buf[0]));
2270         if (ret == 0 && copy_to_user(&uarg->key, &args.key, sizeof(args.key)))
2271                 ret = -EFAULT;
2272         else if (ret == -EOVERFLOW &&
2273                 copy_to_user(&uarg->buf_size, &buf_size, sizeof(buf_size)))
2274                 ret = -EFAULT;
2275
2276         return ret;
2277 }
2278
2279 /*
2280  * Search INODE_REFs to identify path name of 'dirid' directory
2281  * in a 'tree_id' tree. and sets path name to 'name'.
2282  */
2283 static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info,
2284                                 u64 tree_id, u64 dirid, char *name)
2285 {
2286         struct btrfs_root *root;
2287         struct btrfs_key key;
2288         char *ptr;
2289         int ret = -1;
2290         int slot;
2291         int len;
2292         int total_len = 0;
2293         struct btrfs_inode_ref *iref;
2294         struct extent_buffer *l;
2295         struct btrfs_path *path;
2296
2297         if (dirid == BTRFS_FIRST_FREE_OBJECTID) {
2298                 name[0]='\0';
2299                 return 0;
2300         }
2301
2302         path = btrfs_alloc_path();
2303         if (!path)
2304                 return -ENOMEM;
2305
2306         ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX - 1];
2307
2308         root = btrfs_get_fs_root(info, tree_id, true);
2309         if (IS_ERR(root)) {
2310                 ret = PTR_ERR(root);
2311                 root = NULL;
2312                 goto out;
2313         }
2314
2315         key.objectid = dirid;
2316         key.type = BTRFS_INODE_REF_KEY;
2317         key.offset = (u64)-1;
2318
2319         while (1) {
2320                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2321                 if (ret < 0)
2322                         goto out;
2323                 else if (ret > 0) {
2324                         ret = btrfs_previous_item(root, path, dirid,
2325                                                   BTRFS_INODE_REF_KEY);
2326                         if (ret < 0)
2327                                 goto out;
2328                         else if (ret > 0) {
2329                                 ret = -ENOENT;
2330                                 goto out;
2331                         }
2332                 }
2333
2334                 l = path->nodes[0];
2335                 slot = path->slots[0];
2336                 btrfs_item_key_to_cpu(l, &key, slot);
2337
2338                 iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref);
2339                 len = btrfs_inode_ref_name_len(l, iref);
2340                 ptr -= len + 1;
2341                 total_len += len + 1;
2342                 if (ptr < name) {
2343                         ret = -ENAMETOOLONG;
2344                         goto out;
2345                 }
2346
2347                 *(ptr + len) = '/';
2348                 read_extent_buffer(l, ptr, (unsigned long)(iref + 1), len);
2349
2350                 if (key.offset == BTRFS_FIRST_FREE_OBJECTID)
2351                         break;
2352
2353                 btrfs_release_path(path);
2354                 key.objectid = key.offset;
2355                 key.offset = (u64)-1;
2356                 dirid = key.objectid;
2357         }
2358         memmove(name, ptr, total_len);
2359         name[total_len] = '\0';
2360         ret = 0;
2361 out:
2362         btrfs_put_root(root);
2363         btrfs_free_path(path);
2364         return ret;
2365 }
2366
2367 static int btrfs_search_path_in_tree_user(struct inode *inode,
2368                                 struct btrfs_ioctl_ino_lookup_user_args *args)
2369 {
2370         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2371         struct super_block *sb = inode->i_sb;
2372         struct btrfs_key upper_limit = BTRFS_I(inode)->location;
2373         u64 treeid = BTRFS_I(inode)->root->root_key.objectid;
2374         u64 dirid = args->dirid;
2375         unsigned long item_off;
2376         unsigned long item_len;
2377         struct btrfs_inode_ref *iref;
2378         struct btrfs_root_ref *rref;
2379         struct btrfs_root *root = NULL;
2380         struct btrfs_path *path;
2381         struct btrfs_key key, key2;
2382         struct extent_buffer *leaf;
2383         struct inode *temp_inode;
2384         char *ptr;
2385         int slot;
2386         int len;
2387         int total_len = 0;
2388         int ret;
2389
2390         path = btrfs_alloc_path();
2391         if (!path)
2392                 return -ENOMEM;
2393
2394         /*
2395          * If the bottom subvolume does not exist directly under upper_limit,
2396          * construct the path in from the bottom up.
2397          */
2398         if (dirid != upper_limit.objectid) {
2399                 ptr = &args->path[BTRFS_INO_LOOKUP_USER_PATH_MAX - 1];
2400
2401                 root = btrfs_get_fs_root(fs_info, treeid, true);
2402                 if (IS_ERR(root)) {
2403                         ret = PTR_ERR(root);
2404                         goto out;
2405                 }
2406
2407                 key.objectid = dirid;
2408                 key.type = BTRFS_INODE_REF_KEY;
2409                 key.offset = (u64)-1;
2410                 while (1) {
2411                         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2412                         if (ret < 0) {
2413                                 goto out_put;
2414                         } else if (ret > 0) {
2415                                 ret = btrfs_previous_item(root, path, dirid,
2416                                                           BTRFS_INODE_REF_KEY);
2417                                 if (ret < 0) {
2418                                         goto out_put;
2419                                 } else if (ret > 0) {
2420                                         ret = -ENOENT;
2421                                         goto out_put;
2422                                 }
2423                         }
2424
2425                         leaf = path->nodes[0];
2426                         slot = path->slots[0];
2427                         btrfs_item_key_to_cpu(leaf, &key, slot);
2428
2429                         iref = btrfs_item_ptr(leaf, slot, struct btrfs_inode_ref);
2430                         len = btrfs_inode_ref_name_len(leaf, iref);
2431                         ptr -= len + 1;
2432                         total_len += len + 1;
2433                         if (ptr < args->path) {
2434                                 ret = -ENAMETOOLONG;
2435                                 goto out_put;
2436                         }
2437
2438                         *(ptr + len) = '/';
2439                         read_extent_buffer(leaf, ptr,
2440                                         (unsigned long)(iref + 1), len);
2441
2442                         /* Check the read+exec permission of this directory */
2443                         ret = btrfs_previous_item(root, path, dirid,
2444                                                   BTRFS_INODE_ITEM_KEY);
2445                         if (ret < 0) {
2446                                 goto out_put;
2447                         } else if (ret > 0) {
2448                                 ret = -ENOENT;
2449                                 goto out_put;
2450                         }
2451
2452                         leaf = path->nodes[0];
2453                         slot = path->slots[0];
2454                         btrfs_item_key_to_cpu(leaf, &key2, slot);
2455                         if (key2.objectid != dirid) {
2456                                 ret = -ENOENT;
2457                                 goto out_put;
2458                         }
2459
2460                         temp_inode = btrfs_iget(sb, key2.objectid, root);
2461                         if (IS_ERR(temp_inode)) {
2462                                 ret = PTR_ERR(temp_inode);
2463                                 goto out_put;
2464                         }
2465                         ret = inode_permission(temp_inode, MAY_READ | MAY_EXEC);
2466                         iput(temp_inode);
2467                         if (ret) {
2468                                 ret = -EACCES;
2469                                 goto out_put;
2470                         }
2471
2472                         if (key.offset == upper_limit.objectid)
2473                                 break;
2474                         if (key.objectid == BTRFS_FIRST_FREE_OBJECTID) {
2475                                 ret = -EACCES;
2476                                 goto out_put;
2477                         }
2478
2479                         btrfs_release_path(path);
2480                         key.objectid = key.offset;
2481                         key.offset = (u64)-1;
2482                         dirid = key.objectid;
2483                 }
2484
2485                 memmove(args->path, ptr, total_len);
2486                 args->path[total_len] = '\0';
2487                 btrfs_put_root(root);
2488                 root = NULL;
2489                 btrfs_release_path(path);
2490         }
2491
2492         /* Get the bottom subvolume's name from ROOT_REF */
2493         key.objectid = treeid;
2494         key.type = BTRFS_ROOT_REF_KEY;
2495         key.offset = args->treeid;
2496         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
2497         if (ret < 0) {
2498                 goto out;
2499         } else if (ret > 0) {
2500                 ret = -ENOENT;
2501                 goto out;
2502         }
2503
2504         leaf = path->nodes[0];
2505         slot = path->slots[0];
2506         btrfs_item_key_to_cpu(leaf, &key, slot);
2507
2508         item_off = btrfs_item_ptr_offset(leaf, slot);
2509         item_len = btrfs_item_size_nr(leaf, slot);
2510         /* Check if dirid in ROOT_REF corresponds to passed dirid */
2511         rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2512         if (args->dirid != btrfs_root_ref_dirid(leaf, rref)) {
2513                 ret = -EINVAL;
2514                 goto out;
2515         }
2516
2517         /* Copy subvolume's name */
2518         item_off += sizeof(struct btrfs_root_ref);
2519         item_len -= sizeof(struct btrfs_root_ref);
2520         read_extent_buffer(leaf, args->name, item_off, item_len);
2521         args->name[item_len] = 0;
2522
2523 out_put:
2524         btrfs_put_root(root);
2525 out:
2526         btrfs_free_path(path);
2527         return ret;
2528 }
2529
2530 static noinline int btrfs_ioctl_ino_lookup(struct file *file,
2531                                            void __user *argp)
2532 {
2533         struct btrfs_ioctl_ino_lookup_args *args;
2534         struct inode *inode;
2535         int ret = 0;
2536
2537         args = memdup_user(argp, sizeof(*args));
2538         if (IS_ERR(args))
2539                 return PTR_ERR(args);
2540
2541         inode = file_inode(file);
2542
2543         /*
2544          * Unprivileged query to obtain the containing subvolume root id. The
2545          * path is reset so it's consistent with btrfs_search_path_in_tree.
2546          */
2547         if (args->treeid == 0)
2548                 args->treeid = BTRFS_I(inode)->root->root_key.objectid;
2549
2550         if (args->objectid == BTRFS_FIRST_FREE_OBJECTID) {
2551                 args->name[0] = 0;
2552                 goto out;
2553         }
2554
2555         if (!capable(CAP_SYS_ADMIN)) {
2556                 ret = -EPERM;
2557                 goto out;
2558         }
2559
2560         ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info,
2561                                         args->treeid, args->objectid,
2562                                         args->name);
2563
2564 out:
2565         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2566                 ret = -EFAULT;
2567
2568         kfree(args);
2569         return ret;
2570 }
2571
2572 /*
2573  * Version of ino_lookup ioctl (unprivileged)
2574  *
2575  * The main differences from ino_lookup ioctl are:
2576  *
2577  *   1. Read + Exec permission will be checked using inode_permission() during
2578  *      path construction. -EACCES will be returned in case of failure.
2579  *   2. Path construction will be stopped at the inode number which corresponds
2580  *      to the fd with which this ioctl is called. If constructed path does not
2581  *      exist under fd's inode, -EACCES will be returned.
2582  *   3. The name of bottom subvolume is also searched and filled.
2583  */
2584 static int btrfs_ioctl_ino_lookup_user(struct file *file, void __user *argp)
2585 {
2586         struct btrfs_ioctl_ino_lookup_user_args *args;
2587         struct inode *inode;
2588         int ret;
2589
2590         args = memdup_user(argp, sizeof(*args));
2591         if (IS_ERR(args))
2592                 return PTR_ERR(args);
2593
2594         inode = file_inode(file);
2595
2596         if (args->dirid == BTRFS_FIRST_FREE_OBJECTID &&
2597             BTRFS_I(inode)->location.objectid != BTRFS_FIRST_FREE_OBJECTID) {
2598                 /*
2599                  * The subvolume does not exist under fd with which this is
2600                  * called
2601                  */
2602                 kfree(args);
2603                 return -EACCES;
2604         }
2605
2606         ret = btrfs_search_path_in_tree_user(inode, args);
2607
2608         if (ret == 0 && copy_to_user(argp, args, sizeof(*args)))
2609                 ret = -EFAULT;
2610
2611         kfree(args);
2612         return ret;
2613 }
2614
2615 /* Get the subvolume information in BTRFS_ROOT_ITEM and BTRFS_ROOT_BACKREF */
2616 static int btrfs_ioctl_get_subvol_info(struct file *file, void __user *argp)
2617 {
2618         struct btrfs_ioctl_get_subvol_info_args *subvol_info;
2619         struct btrfs_fs_info *fs_info;
2620         struct btrfs_root *root;
2621         struct btrfs_path *path;
2622         struct btrfs_key key;
2623         struct btrfs_root_item *root_item;
2624         struct btrfs_root_ref *rref;
2625         struct extent_buffer *leaf;
2626         unsigned long item_off;
2627         unsigned long item_len;
2628         struct inode *inode;
2629         int slot;
2630         int ret = 0;
2631
2632         path = btrfs_alloc_path();
2633         if (!path)
2634                 return -ENOMEM;
2635
2636         subvol_info = kzalloc(sizeof(*subvol_info), GFP_KERNEL);
2637         if (!subvol_info) {
2638                 btrfs_free_path(path);
2639                 return -ENOMEM;
2640         }
2641
2642         inode = file_inode(file);
2643         fs_info = BTRFS_I(inode)->root->fs_info;
2644
2645         /* Get root_item of inode's subvolume */
2646         key.objectid = BTRFS_I(inode)->root->root_key.objectid;
2647         root = btrfs_get_fs_root(fs_info, key.objectid, true);
2648         if (IS_ERR(root)) {
2649                 ret = PTR_ERR(root);
2650                 goto out_free;
2651         }
2652         root_item = &root->root_item;
2653
2654         subvol_info->treeid = key.objectid;
2655
2656         subvol_info->generation = btrfs_root_generation(root_item);
2657         subvol_info->flags = btrfs_root_flags(root_item);
2658
2659         memcpy(subvol_info->uuid, root_item->uuid, BTRFS_UUID_SIZE);
2660         memcpy(subvol_info->parent_uuid, root_item->parent_uuid,
2661                                                     BTRFS_UUID_SIZE);
2662         memcpy(subvol_info->received_uuid, root_item->received_uuid,
2663                                                     BTRFS_UUID_SIZE);
2664
2665         subvol_info->ctransid = btrfs_root_ctransid(root_item);
2666         subvol_info->ctime.sec = btrfs_stack_timespec_sec(&root_item->ctime);
2667         subvol_info->ctime.nsec = btrfs_stack_timespec_nsec(&root_item->ctime);
2668
2669         subvol_info->otransid = btrfs_root_otransid(root_item);
2670         subvol_info->otime.sec = btrfs_stack_timespec_sec(&root_item->otime);
2671         subvol_info->otime.nsec = btrfs_stack_timespec_nsec(&root_item->otime);
2672
2673         subvol_info->stransid = btrfs_root_stransid(root_item);
2674         subvol_info->stime.sec = btrfs_stack_timespec_sec(&root_item->stime);
2675         subvol_info->stime.nsec = btrfs_stack_timespec_nsec(&root_item->stime);
2676
2677         subvol_info->rtransid = btrfs_root_rtransid(root_item);
2678         subvol_info->rtime.sec = btrfs_stack_timespec_sec(&root_item->rtime);
2679         subvol_info->rtime.nsec = btrfs_stack_timespec_nsec(&root_item->rtime);
2680
2681         if (key.objectid != BTRFS_FS_TREE_OBJECTID) {
2682                 /* Search root tree for ROOT_BACKREF of this subvolume */
2683                 key.type = BTRFS_ROOT_BACKREF_KEY;
2684                 key.offset = 0;
2685                 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
2686                 if (ret < 0) {
2687                         goto out;
2688                 } else if (path->slots[0] >=
2689                            btrfs_header_nritems(path->nodes[0])) {
2690                         ret = btrfs_next_leaf(fs_info->tree_root, path);
2691                         if (ret < 0) {
2692                                 goto out;
2693                         } else if (ret > 0) {
2694                                 ret = -EUCLEAN;
2695                                 goto out;
2696                         }
2697                 }
2698
2699                 leaf = path->nodes[0];
2700                 slot = path->slots[0];
2701                 btrfs_item_key_to_cpu(leaf, &key, slot);
2702                 if (key.objectid == subvol_info->treeid &&
2703                     key.type == BTRFS_ROOT_BACKREF_KEY) {
2704                         subvol_info->parent_id = key.offset;
2705
2706                         rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2707                         subvol_info->dirid = btrfs_root_ref_dirid(leaf, rref);
2708
2709                         item_off = btrfs_item_ptr_offset(leaf, slot)
2710                                         + sizeof(struct btrfs_root_ref);
2711                         item_len = btrfs_item_size_nr(leaf, slot)
2712                                         - sizeof(struct btrfs_root_ref);
2713                         read_extent_buffer(leaf, subvol_info->name,
2714                                            item_off, item_len);
2715                 } else {
2716                         ret = -ENOENT;
2717                         goto out;
2718                 }
2719         }
2720
2721         if (copy_to_user(argp, subvol_info, sizeof(*subvol_info)))
2722                 ret = -EFAULT;
2723
2724 out:
2725         btrfs_put_root(root);
2726 out_free:
2727         btrfs_free_path(path);
2728         kfree(subvol_info);
2729         return ret;
2730 }
2731
2732 /*
2733  * Return ROOT_REF information of the subvolume containing this inode
2734  * except the subvolume name.
2735  */
2736 static int btrfs_ioctl_get_subvol_rootref(struct file *file, void __user *argp)
2737 {
2738         struct btrfs_ioctl_get_subvol_rootref_args *rootrefs;
2739         struct btrfs_root_ref *rref;
2740         struct btrfs_root *root;
2741         struct btrfs_path *path;
2742         struct btrfs_key key;
2743         struct extent_buffer *leaf;
2744         struct inode *inode;
2745         u64 objectid;
2746         int slot;
2747         int ret;
2748         u8 found;
2749
2750         path = btrfs_alloc_path();
2751         if (!path)
2752                 return -ENOMEM;
2753
2754         rootrefs = memdup_user(argp, sizeof(*rootrefs));
2755         if (IS_ERR(rootrefs)) {
2756                 btrfs_free_path(path);
2757                 return PTR_ERR(rootrefs);
2758         }
2759
2760         inode = file_inode(file);
2761         root = BTRFS_I(inode)->root->fs_info->tree_root;
2762         objectid = BTRFS_I(inode)->root->root_key.objectid;
2763
2764         key.objectid = objectid;
2765         key.type = BTRFS_ROOT_REF_KEY;
2766         key.offset = rootrefs->min_treeid;
2767         found = 0;
2768
2769         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2770         if (ret < 0) {
2771                 goto out;
2772         } else if (path->slots[0] >=
2773                    btrfs_header_nritems(path->nodes[0])) {
2774                 ret = btrfs_next_leaf(root, path);
2775                 if (ret < 0) {
2776                         goto out;
2777                 } else if (ret > 0) {
2778                         ret = -EUCLEAN;
2779                         goto out;
2780                 }
2781         }
2782         while (1) {
2783                 leaf = path->nodes[0];
2784                 slot = path->slots[0];
2785
2786                 btrfs_item_key_to_cpu(leaf, &key, slot);
2787                 if (key.objectid != objectid || key.type != BTRFS_ROOT_REF_KEY) {
2788                         ret = 0;
2789                         goto out;
2790                 }
2791
2792                 if (found == BTRFS_MAX_ROOTREF_BUFFER_NUM) {
2793                         ret = -EOVERFLOW;
2794                         goto out;
2795                 }
2796
2797                 rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref);
2798                 rootrefs->rootref[found].treeid = key.offset;
2799                 rootrefs->rootref[found].dirid =
2800                                   btrfs_root_ref_dirid(leaf, rref);
2801                 found++;
2802
2803                 ret = btrfs_next_item(root, path);
2804                 if (ret < 0) {
2805                         goto out;
2806                 } else if (ret > 0) {
2807                         ret = -EUCLEAN;
2808                         goto out;
2809                 }
2810         }
2811
2812 out:
2813         if (!ret || ret == -EOVERFLOW) {
2814                 rootrefs->num_items = found;
2815                 /* update min_treeid for next search */
2816                 if (found)
2817                         rootrefs->min_treeid =
2818                                 rootrefs->rootref[found - 1].treeid + 1;
2819                 if (copy_to_user(argp, rootrefs, sizeof(*rootrefs)))
2820                         ret = -EFAULT;
2821         }
2822
2823         kfree(rootrefs);
2824         btrfs_free_path(path);
2825
2826         return ret;
2827 }
2828
2829 static noinline int btrfs_ioctl_snap_destroy(struct file *file,
2830                                              void __user *arg,
2831                                              bool destroy_v2)
2832 {
2833         struct dentry *parent = file->f_path.dentry;
2834         struct btrfs_fs_info *fs_info = btrfs_sb(parent->d_sb);
2835         struct dentry *dentry;
2836         struct inode *dir = d_inode(parent);
2837         struct inode *inode;
2838         struct btrfs_root *root = BTRFS_I(dir)->root;
2839         struct btrfs_root *dest = NULL;
2840         struct btrfs_ioctl_vol_args *vol_args = NULL;
2841         struct btrfs_ioctl_vol_args_v2 *vol_args2 = NULL;
2842         char *subvol_name, *subvol_name_ptr = NULL;
2843         int subvol_namelen;
2844         int err = 0;
2845         bool destroy_parent = false;
2846
2847         if (destroy_v2) {
2848                 vol_args2 = memdup_user(arg, sizeof(*vol_args2));
2849                 if (IS_ERR(vol_args2))
2850                         return PTR_ERR(vol_args2);
2851
2852                 if (vol_args2->flags & ~BTRFS_SUBVOL_DELETE_ARGS_MASK) {
2853                         err = -EOPNOTSUPP;
2854                         goto out;
2855                 }
2856
2857                 /*
2858                  * If SPEC_BY_ID is not set, we are looking for the subvolume by
2859                  * name, same as v1 currently does.
2860                  */
2861                 if (!(vol_args2->flags & BTRFS_SUBVOL_SPEC_BY_ID)) {
2862                         vol_args2->name[BTRFS_SUBVOL_NAME_MAX] = 0;
2863                         subvol_name = vol_args2->name;
2864
2865                         err = mnt_want_write_file(file);
2866                         if (err)
2867                                 goto out;
2868                 } else {
2869                         if (vol_args2->subvolid < BTRFS_FIRST_FREE_OBJECTID) {
2870                                 err = -EINVAL;
2871                                 goto out;
2872                         }
2873
2874                         err = mnt_want_write_file(file);
2875                         if (err)
2876                                 goto out;
2877
2878                         dentry = btrfs_get_dentry(fs_info->sb,
2879                                         BTRFS_FIRST_FREE_OBJECTID,
2880                                         vol_args2->subvolid, 0, 0);
2881                         if (IS_ERR(dentry)) {
2882                                 err = PTR_ERR(dentry);
2883                                 goto out_drop_write;
2884                         }
2885
2886                         /*
2887                          * Change the default parent since the subvolume being
2888                          * deleted can be outside of the current mount point.
2889                          */
2890                         parent = btrfs_get_parent(dentry);
2891
2892                         /*
2893                          * At this point dentry->d_name can point to '/' if the
2894                          * subvolume we want to destroy is outsite of the
2895                          * current mount point, so we need to release the
2896                          * current dentry and execute the lookup to return a new
2897                          * one with ->d_name pointing to the
2898                          * <mount point>/subvol_name.
2899                          */
2900                         dput(dentry);
2901                         if (IS_ERR(parent)) {
2902                                 err = PTR_ERR(parent);
2903                                 goto out_drop_write;
2904                         }
2905                         dir = d_inode(parent);
2906
2907                         /*
2908                          * If v2 was used with SPEC_BY_ID, a new parent was
2909                          * allocated since the subvolume can be outside of the
2910                          * current mount point. Later on we need to release this
2911                          * new parent dentry.
2912                          */
2913                         destroy_parent = true;
2914
2915                         subvol_name_ptr = btrfs_get_subvol_name_from_objectid(
2916                                                 fs_info, vol_args2->subvolid);
2917                         if (IS_ERR(subvol_name_ptr)) {
2918                                 err = PTR_ERR(subvol_name_ptr);
2919                                 goto free_parent;
2920                         }
2921                         /* subvol_name_ptr is already NULL termined */
2922                         subvol_name = (char *)kbasename(subvol_name_ptr);
2923                 }
2924         } else {
2925                 vol_args = memdup_user(arg, sizeof(*vol_args));
2926                 if (IS_ERR(vol_args))
2927                         return PTR_ERR(vol_args);
2928
2929                 vol_args->name[BTRFS_PATH_NAME_MAX] = 0;
2930                 subvol_name = vol_args->name;
2931
2932                 err = mnt_want_write_file(file);
2933                 if (err)
2934                         goto out;
2935         }
2936
2937         subvol_namelen = strlen(subvol_name);
2938
2939         if (strchr(subvol_name, '/') ||
2940             strncmp(subvol_name, "..", subvol_namelen) == 0) {
2941                 err = -EINVAL;
2942                 goto free_subvol_name;
2943         }
2944
2945         if (!S_ISDIR(dir->i_mode)) {
2946                 err = -ENOTDIR;
2947                 goto free_subvol_name;
2948         }
2949
2950         err = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
2951         if (err == -EINTR)
2952                 goto free_subvol_name;
2953         dentry = lookup_one_len(subvol_name, parent, subvol_namelen);
2954         if (IS_ERR(dentry)) {
2955                 err = PTR_ERR(dentry);
2956                 goto out_unlock_dir;
2957         }
2958
2959         if (d_really_is_negative(dentry)) {
2960                 err = -ENOENT;
2961                 goto out_dput;
2962         }
2963
2964         inode = d_inode(dentry);
2965         dest = BTRFS_I(inode)->root;
2966         if (!capable(CAP_SYS_ADMIN)) {
2967                 /*
2968                  * Regular user.  Only allow this with a special mount
2969                  * option, when the user has write+exec access to the
2970                  * subvol root, and when rmdir(2) would have been
2971                  * allowed.
2972                  *
2973                  * Note that this is _not_ check that the subvol is
2974                  * empty or doesn't contain data that we wouldn't
2975                  * otherwise be able to delete.
2976                  *
2977                  * Users who want to delete empty subvols should try
2978                  * rmdir(2).
2979                  */
2980                 err = -EPERM;
2981                 if (!btrfs_test_opt(fs_info, USER_SUBVOL_RM_ALLOWED))
2982                         goto out_dput;
2983
2984                 /*
2985                  * Do not allow deletion if the parent dir is the same
2986                  * as the dir to be deleted.  That means the ioctl
2987                  * must be called on the dentry referencing the root
2988                  * of the subvol, not a random directory contained
2989                  * within it.
2990                  */
2991                 err = -EINVAL;
2992                 if (root == dest)
2993                         goto out_dput;
2994
2995                 err = inode_permission(inode, MAY_WRITE | MAY_EXEC);
2996                 if (err)
2997                         goto out_dput;
2998         }
2999
3000         /* check if subvolume may be deleted by a user */
3001         err = btrfs_may_delete(dir, dentry, 1);
3002         if (err)
3003                 goto out_dput;
3004
3005         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
3006                 err = -EINVAL;
3007                 goto out_dput;
3008         }
3009
3010         inode_lock(inode);
3011         err = btrfs_delete_subvolume(dir, dentry);
3012         inode_unlock(inode);
3013         if (!err) {
3014                 fsnotify_rmdir(dir, dentry);
3015                 d_delete(dentry);
3016         }
3017
3018 out_dput:
3019         dput(dentry);
3020 out_unlock_dir:
3021         inode_unlock(dir);
3022 free_subvol_name:
3023         kfree(subvol_name_ptr);
3024 free_parent:
3025         if (destroy_parent)
3026                 dput(parent);
3027 out_drop_write:
3028         mnt_drop_write_file(file);
3029 out:
3030         kfree(vol_args2);
3031         kfree(vol_args);
3032         return err;
3033 }
3034
3035 static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
3036 {
3037         struct inode *inode = file_inode(file);
3038         struct btrfs_root *root = BTRFS_I(inode)->root;
3039         struct btrfs_ioctl_defrag_range_args *range;
3040         int ret;
3041
3042         ret = mnt_want_write_file(file);
3043         if (ret)
3044                 return ret;
3045
3046         if (btrfs_root_readonly(root)) {
3047                 ret = -EROFS;
3048                 goto out;
3049         }
3050
3051         switch (inode->i_mode & S_IFMT) {
3052         case S_IFDIR:
3053                 if (!capable(CAP_SYS_ADMIN)) {
3054                         ret = -EPERM;
3055                         goto out;
3056                 }
3057                 ret = btrfs_defrag_root(root);
3058                 break;
3059         case S_IFREG:
3060                 /*
3061                  * Note that this does not check the file descriptor for write
3062                  * access. This prevents defragmenting executables that are
3063                  * running and allows defrag on files open in read-only mode.
3064                  */
3065                 if (!capable(CAP_SYS_ADMIN) &&
3066                     inode_permission(inode, MAY_WRITE)) {
3067                         ret = -EPERM;
3068                         goto out;
3069                 }
3070
3071                 range = kzalloc(sizeof(*range), GFP_KERNEL);
3072                 if (!range) {
3073                         ret = -ENOMEM;
3074                         goto out;
3075                 }
3076
3077                 if (argp) {
3078                         if (copy_from_user(range, argp,
3079                                            sizeof(*range))) {
3080                                 ret = -EFAULT;
3081                                 kfree(range);
3082                                 goto out;
3083                         }
3084                         /* compression requires us to start the IO */
3085                         if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) {
3086                                 range->flags |= BTRFS_DEFRAG_RANGE_START_IO;
3087                                 range->extent_thresh = (u32)-1;
3088                         }
3089                 } else {
3090                         /* the rest are all set to zero by kzalloc */
3091                         range->len = (u64)-1;
3092                 }
3093                 ret = btrfs_defrag_file(file_inode(file), file,
3094                                         range, BTRFS_OLDEST_GENERATION, 0);
3095                 if (ret > 0)
3096                         ret = 0;
3097                 kfree(range);
3098                 break;
3099         default:
3100                 ret = -EINVAL;
3101         }
3102 out:
3103         mnt_drop_write_file(file);
3104         return ret;
3105 }
3106
3107 static long btrfs_ioctl_add_dev(struct btrfs_fs_info *fs_info, void __user *arg)
3108 {
3109         struct btrfs_ioctl_vol_args *vol_args;
3110         int ret;
3111
3112         if (!capable(CAP_SYS_ADMIN))
3113                 return -EPERM;
3114
3115         if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags))
3116                 return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3117
3118         vol_args = memdup_user(arg, sizeof(*vol_args));
3119         if (IS_ERR(vol_args)) {
3120                 ret = PTR_ERR(vol_args);
3121                 goto out;
3122         }
3123
3124         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
3125         ret = btrfs_init_new_device(fs_info, vol_args->name);
3126
3127         if (!ret)
3128                 btrfs_info(fs_info, "disk added %s", vol_args->name);
3129
3130         kfree(vol_args);
3131 out:
3132         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
3133         return ret;
3134 }
3135
3136 static long btrfs_ioctl_rm_dev_v2(struct file *file, void __user *arg)
3137 {
3138         struct inode *inode = file_inode(file);
3139         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3140         struct btrfs_ioctl_vol_args_v2 *vol_args;
3141         int ret;
3142
3143         if (!capable(CAP_SYS_ADMIN))
3144                 return -EPERM;
3145
3146         ret = mnt_want_write_file(file);
3147         if (ret)
3148                 return ret;
3149
3150         vol_args = memdup_user(arg, sizeof(*vol_args));
3151         if (IS_ERR(vol_args)) {
3152                 ret = PTR_ERR(vol_args);
3153                 goto err_drop;
3154         }
3155
3156         if (vol_args->flags & ~BTRFS_DEVICE_REMOVE_ARGS_MASK) {
3157                 ret = -EOPNOTSUPP;
3158                 goto out;
3159         }
3160
3161         if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
3162                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3163                 goto out;
3164         }
3165
3166         if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID) {
3167                 ret = btrfs_rm_device(fs_info, NULL, vol_args->devid);
3168         } else {
3169                 vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0';
3170                 ret = btrfs_rm_device(fs_info, vol_args->name, 0);
3171         }
3172         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
3173
3174         if (!ret) {
3175                 if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID)
3176                         btrfs_info(fs_info, "device deleted: id %llu",
3177                                         vol_args->devid);
3178                 else
3179                         btrfs_info(fs_info, "device deleted: %s",
3180                                         vol_args->name);
3181         }
3182 out:
3183         kfree(vol_args);
3184 err_drop:
3185         mnt_drop_write_file(file);
3186         return ret;
3187 }
3188
3189 static long btrfs_ioctl_rm_dev(struct file *file, void __user *arg)
3190 {
3191         struct inode *inode = file_inode(file);
3192         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3193         struct btrfs_ioctl_vol_args *vol_args;
3194         int ret;
3195
3196         if (!capable(CAP_SYS_ADMIN))
3197                 return -EPERM;
3198
3199         ret = mnt_want_write_file(file);
3200         if (ret)
3201                 return ret;
3202
3203         if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
3204                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3205                 goto out_drop_write;
3206         }
3207
3208         vol_args = memdup_user(arg, sizeof(*vol_args));
3209         if (IS_ERR(vol_args)) {
3210                 ret = PTR_ERR(vol_args);
3211                 goto out;
3212         }
3213
3214         vol_args->name[BTRFS_PATH_NAME_MAX] = '\0';
3215         ret = btrfs_rm_device(fs_info, vol_args->name, 0);
3216
3217         if (!ret)
3218                 btrfs_info(fs_info, "disk deleted %s", vol_args->name);
3219         kfree(vol_args);
3220 out:
3221         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
3222 out_drop_write:
3223         mnt_drop_write_file(file);
3224
3225         return ret;
3226 }
3227
3228 static long btrfs_ioctl_fs_info(struct btrfs_fs_info *fs_info,
3229                                 void __user *arg)
3230 {
3231         struct btrfs_ioctl_fs_info_args *fi_args;
3232         struct btrfs_device *device;
3233         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
3234         u64 flags_in;
3235         int ret = 0;
3236
3237         fi_args = memdup_user(arg, sizeof(*fi_args));
3238         if (IS_ERR(fi_args))
3239                 return PTR_ERR(fi_args);
3240
3241         flags_in = fi_args->flags;
3242         memset(fi_args, 0, sizeof(*fi_args));
3243
3244         rcu_read_lock();
3245         fi_args->num_devices = fs_devices->num_devices;
3246
3247         list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
3248                 if (device->devid > fi_args->max_id)
3249                         fi_args->max_id = device->devid;
3250         }
3251         rcu_read_unlock();
3252
3253         memcpy(&fi_args->fsid, fs_devices->fsid, sizeof(fi_args->fsid));
3254         fi_args->nodesize = fs_info->nodesize;
3255         fi_args->sectorsize = fs_info->sectorsize;
3256         fi_args->clone_alignment = fs_info->sectorsize;
3257
3258         if (flags_in & BTRFS_FS_INFO_FLAG_CSUM_INFO) {
3259                 fi_args->csum_type = btrfs_super_csum_type(fs_info->super_copy);
3260                 fi_args->csum_size = btrfs_super_csum_size(fs_info->super_copy);
3261                 fi_args->flags |= BTRFS_FS_INFO_FLAG_CSUM_INFO;
3262         }
3263
3264         if (flags_in & BTRFS_FS_INFO_FLAG_GENERATION) {
3265                 fi_args->generation = fs_info->generation;
3266                 fi_args->flags |= BTRFS_FS_INFO_FLAG_GENERATION;
3267         }
3268
3269         if (flags_in & BTRFS_FS_INFO_FLAG_METADATA_UUID) {
3270                 memcpy(&fi_args->metadata_uuid, fs_devices->metadata_uuid,
3271                        sizeof(fi_args->metadata_uuid));
3272                 fi_args->flags |= BTRFS_FS_INFO_FLAG_METADATA_UUID;
3273         }
3274
3275         if (copy_to_user(arg, fi_args, sizeof(*fi_args)))
3276                 ret = -EFAULT;
3277
3278         kfree(fi_args);
3279         return ret;
3280 }
3281
3282 static long btrfs_ioctl_dev_info(struct btrfs_fs_info *fs_info,
3283                                  void __user *arg)
3284 {
3285         struct btrfs_ioctl_dev_info_args *di_args;
3286         struct btrfs_device *dev;
3287         int ret = 0;
3288         char *s_uuid = NULL;
3289
3290         di_args = memdup_user(arg, sizeof(*di_args));
3291         if (IS_ERR(di_args))
3292                 return PTR_ERR(di_args);
3293
3294         if (!btrfs_is_empty_uuid(di_args->uuid))
3295                 s_uuid = di_args->uuid;
3296
3297         rcu_read_lock();
3298         dev = btrfs_find_device(fs_info->fs_devices, di_args->devid, s_uuid,
3299                                 NULL, true);
3300
3301         if (!dev) {
3302                 ret = -ENODEV;
3303                 goto out;
3304         }
3305
3306         di_args->devid = dev->devid;
3307         di_args->bytes_used = btrfs_device_get_bytes_used(dev);
3308         di_args->total_bytes = btrfs_device_get_total_bytes(dev);
3309         memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid));
3310         if (dev->name) {
3311                 strncpy(di_args->path, rcu_str_deref(dev->name),
3312                                 sizeof(di_args->path) - 1);
3313                 di_args->path[sizeof(di_args->path) - 1] = 0;
3314         } else {
3315                 di_args->path[0] = '\0';
3316         }
3317
3318 out:
3319         rcu_read_unlock();
3320         if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args)))
3321                 ret = -EFAULT;
3322
3323         kfree(di_args);
3324         return ret;
3325 }
3326
3327 static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp)
3328 {
3329         struct inode *inode = file_inode(file);
3330         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3331         struct btrfs_root *root = BTRFS_I(inode)->root;
3332         struct btrfs_root *new_root;
3333         struct btrfs_dir_item *di;
3334         struct btrfs_trans_handle *trans;
3335         struct btrfs_path *path = NULL;
3336         struct btrfs_disk_key disk_key;
3337         u64 objectid = 0;
3338         u64 dir_id;
3339         int ret;
3340
3341         if (!capable(CAP_SYS_ADMIN))
3342                 return -EPERM;
3343
3344         ret = mnt_want_write_file(file);
3345         if (ret)
3346                 return ret;
3347
3348         if (copy_from_user(&objectid, argp, sizeof(objectid))) {
3349                 ret = -EFAULT;
3350                 goto out;
3351         }
3352
3353         if (!objectid)
3354                 objectid = BTRFS_FS_TREE_OBJECTID;
3355
3356         new_root = btrfs_get_fs_root(fs_info, objectid, true);
3357         if (IS_ERR(new_root)) {
3358                 ret = PTR_ERR(new_root);
3359                 goto out;
3360         }
3361         if (!is_fstree(new_root->root_key.objectid)) {
3362                 ret = -ENOENT;
3363                 goto out_free;
3364         }
3365
3366         path = btrfs_alloc_path();
3367         if (!path) {
3368                 ret = -ENOMEM;
3369                 goto out_free;
3370         }
3371         path->leave_spinning = 1;
3372
3373         trans = btrfs_start_transaction(root, 1);
3374         if (IS_ERR(trans)) {
3375                 ret = PTR_ERR(trans);
3376                 goto out_free;
3377         }
3378
3379         dir_id = btrfs_super_root_dir(fs_info->super_copy);
3380         di = btrfs_lookup_dir_item(trans, fs_info->tree_root, path,
3381                                    dir_id, "default", 7, 1);
3382         if (IS_ERR_OR_NULL(di)) {
3383                 btrfs_release_path(path);
3384                 btrfs_end_transaction(trans);
3385                 btrfs_err(fs_info,
3386                           "Umm, you don't have the default diritem, this isn't going to work");
3387                 ret = -ENOENT;
3388                 goto out_free;
3389         }
3390
3391         btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key);
3392         btrfs_set_dir_item_key(path->nodes[0], di, &disk_key);
3393         btrfs_mark_buffer_dirty(path->nodes[0]);
3394         btrfs_release_path(path);
3395
3396         btrfs_set_fs_incompat(fs_info, DEFAULT_SUBVOL);
3397         btrfs_end_transaction(trans);
3398 out_free:
3399         btrfs_put_root(new_root);
3400         btrfs_free_path(path);
3401 out:
3402         mnt_drop_write_file(file);
3403         return ret;
3404 }
3405
3406 static void get_block_group_info(struct list_head *groups_list,
3407                                  struct btrfs_ioctl_space_info *space)
3408 {
3409         struct btrfs_block_group *block_group;
3410
3411         space->total_bytes = 0;
3412         space->used_bytes = 0;
3413         space->flags = 0;
3414         list_for_each_entry(block_group, groups_list, list) {
3415                 space->flags = block_group->flags;
3416                 space->total_bytes += block_group->length;
3417                 space->used_bytes += block_group->used;
3418         }
3419 }
3420
3421 static long btrfs_ioctl_space_info(struct btrfs_fs_info *fs_info,
3422                                    void __user *arg)
3423 {
3424         struct btrfs_ioctl_space_args space_args;
3425         struct btrfs_ioctl_space_info space;
3426         struct btrfs_ioctl_space_info *dest;
3427         struct btrfs_ioctl_space_info *dest_orig;
3428         struct btrfs_ioctl_space_info __user *user_dest;
3429         struct btrfs_space_info *info;
3430         static const u64 types[] = {
3431                 BTRFS_BLOCK_GROUP_DATA,
3432                 BTRFS_BLOCK_GROUP_SYSTEM,
3433                 BTRFS_BLOCK_GROUP_METADATA,
3434                 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA
3435         };
3436         int num_types = 4;
3437         int alloc_size;
3438         int ret = 0;
3439         u64 slot_count = 0;
3440         int i, c;
3441
3442         if (copy_from_user(&space_args,
3443                            (struct btrfs_ioctl_space_args __user *)arg,
3444                            sizeof(space_args)))
3445                 return -EFAULT;
3446
3447         for (i = 0; i < num_types; i++) {
3448                 struct btrfs_space_info *tmp;
3449
3450                 info = NULL;
3451                 rcu_read_lock();
3452                 list_for_each_entry_rcu(tmp, &fs_info->space_info,
3453                                         list) {
3454                         if (tmp->flags == types[i]) {
3455                                 info = tmp;
3456                                 break;
3457                         }
3458                 }
3459                 rcu_read_unlock();
3460
3461                 if (!info)
3462                         continue;
3463
3464                 down_read(&info->groups_sem);
3465                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3466                         if (!list_empty(&info->block_groups[c]))
3467                                 slot_count++;
3468                 }
3469                 up_read(&info->groups_sem);
3470         }
3471
3472         /*
3473          * Global block reserve, exported as a space_info
3474          */
3475         slot_count++;
3476
3477         /* space_slots == 0 means they are asking for a count */
3478         if (space_args.space_slots == 0) {
3479                 space_args.total_spaces = slot_count;
3480                 goto out;
3481         }
3482
3483         slot_count = min_t(u64, space_args.space_slots, slot_count);
3484
3485         alloc_size = sizeof(*dest) * slot_count;
3486
3487         /* we generally have at most 6 or so space infos, one for each raid
3488          * level.  So, a whole page should be more than enough for everyone
3489          */
3490         if (alloc_size > PAGE_SIZE)
3491                 return -ENOMEM;
3492
3493         space_args.total_spaces = 0;
3494         dest = kmalloc(alloc_size, GFP_KERNEL);
3495         if (!dest)
3496                 return -ENOMEM;
3497         dest_orig = dest;
3498
3499         /* now we have a buffer to copy into */
3500         for (i = 0; i < num_types; i++) {
3501                 struct btrfs_space_info *tmp;
3502
3503                 if (!slot_count)
3504                         break;
3505
3506                 info = NULL;
3507                 rcu_read_lock();
3508                 list_for_each_entry_rcu(tmp, &fs_info->space_info,
3509                                         list) {
3510                         if (tmp->flags == types[i]) {
3511                                 info = tmp;
3512                                 break;
3513                         }
3514                 }
3515                 rcu_read_unlock();
3516
3517                 if (!info)
3518                         continue;
3519                 down_read(&info->groups_sem);
3520                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3521                         if (!list_empty(&info->block_groups[c])) {
3522                                 get_block_group_info(&info->block_groups[c],
3523                                                      &space);
3524                                 memcpy(dest, &space, sizeof(space));
3525                                 dest++;
3526                                 space_args.total_spaces++;
3527                                 slot_count--;
3528                         }
3529                         if (!slot_count)
3530                                 break;
3531                 }
3532                 up_read(&info->groups_sem);
3533         }
3534
3535         /*
3536          * Add global block reserve
3537          */
3538         if (slot_count) {
3539                 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
3540
3541                 spin_lock(&block_rsv->lock);
3542                 space.total_bytes = block_rsv->size;
3543                 space.used_bytes = block_rsv->size - block_rsv->reserved;
3544                 spin_unlock(&block_rsv->lock);
3545                 space.flags = BTRFS_SPACE_INFO_GLOBAL_RSV;
3546                 memcpy(dest, &space, sizeof(space));
3547                 space_args.total_spaces++;
3548         }
3549
3550         user_dest = (struct btrfs_ioctl_space_info __user *)
3551                 (arg + sizeof(struct btrfs_ioctl_space_args));
3552
3553         if (copy_to_user(user_dest, dest_orig, alloc_size))
3554                 ret = -EFAULT;
3555
3556         kfree(dest_orig);
3557 out:
3558         if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args)))
3559                 ret = -EFAULT;
3560
3561         return ret;
3562 }
3563
3564 static noinline long btrfs_ioctl_start_sync(struct btrfs_root *root,
3565                                             void __user *argp)
3566 {
3567         struct btrfs_trans_handle *trans;
3568         u64 transid;
3569         int ret;
3570
3571         trans = btrfs_attach_transaction_barrier(root);
3572         if (IS_ERR(trans)) {
3573                 if (PTR_ERR(trans) != -ENOENT)
3574                         return PTR_ERR(trans);
3575
3576                 /* No running transaction, don't bother */
3577                 transid = root->fs_info->last_trans_committed;
3578                 goto out;
3579         }
3580         transid = trans->transid;
3581         ret = btrfs_commit_transaction_async(trans, 0);
3582         if (ret) {
3583                 btrfs_end_transaction(trans);
3584                 return ret;
3585         }
3586 out:
3587         if (argp)
3588                 if (copy_to_user(argp, &transid, sizeof(transid)))
3589                         return -EFAULT;
3590         return 0;
3591 }
3592
3593 static noinline long btrfs_ioctl_wait_sync(struct btrfs_fs_info *fs_info,
3594                                            void __user *argp)
3595 {
3596         u64 transid;
3597
3598         if (argp) {
3599                 if (copy_from_user(&transid, argp, sizeof(transid)))
3600                         return -EFAULT;
3601         } else {
3602                 transid = 0;  /* current trans */
3603         }
3604         return btrfs_wait_for_commit(fs_info, transid);
3605 }
3606
3607 static long btrfs_ioctl_scrub(struct file *file, void __user *arg)
3608 {
3609         struct btrfs_fs_info *fs_info = btrfs_sb(file_inode(file)->i_sb);
3610         struct btrfs_ioctl_scrub_args *sa;
3611         int ret;
3612
3613         if (!capable(CAP_SYS_ADMIN))
3614                 return -EPERM;
3615
3616         sa = memdup_user(arg, sizeof(*sa));
3617         if (IS_ERR(sa))
3618                 return PTR_ERR(sa);
3619
3620         if (!(sa->flags & BTRFS_SCRUB_READONLY)) {
3621                 ret = mnt_want_write_file(file);
3622                 if (ret)
3623                         goto out;
3624         }
3625
3626         ret = btrfs_scrub_dev(fs_info, sa->devid, sa->start, sa->end,
3627                               &sa->progress, sa->flags & BTRFS_SCRUB_READONLY,
3628                               0);
3629
3630         /*
3631          * Copy scrub args to user space even if btrfs_scrub_dev() returned an
3632          * error. This is important as it allows user space to know how much
3633          * progress scrub has done. For example, if scrub is canceled we get
3634          * -ECANCELED from btrfs_scrub_dev() and return that error back to user
3635          * space. Later user space can inspect the progress from the structure
3636          * btrfs_ioctl_scrub_args and resume scrub from where it left off
3637          * previously (btrfs-progs does this).
3638          * If we fail to copy the btrfs_ioctl_scrub_args structure to user space
3639          * then return -EFAULT to signal the structure was not copied or it may
3640          * be corrupt and unreliable due to a partial copy.
3641          */
3642         if (copy_to_user(arg, sa, sizeof(*sa)))
3643                 ret = -EFAULT;
3644
3645         if (!(sa->flags & BTRFS_SCRUB_READONLY))
3646                 mnt_drop_write_file(file);
3647 out:
3648         kfree(sa);
3649         return ret;
3650 }
3651
3652 static long btrfs_ioctl_scrub_cancel(struct btrfs_fs_info *fs_info)
3653 {
3654         if (!capable(CAP_SYS_ADMIN))
3655                 return -EPERM;
3656
3657         return btrfs_scrub_cancel(fs_info);
3658 }
3659
3660 static long btrfs_ioctl_scrub_progress(struct btrfs_fs_info *fs_info,
3661                                        void __user *arg)
3662 {
3663         struct btrfs_ioctl_scrub_args *sa;
3664         int ret;
3665
3666         if (!capable(CAP_SYS_ADMIN))
3667                 return -EPERM;
3668
3669         sa = memdup_user(arg, sizeof(*sa));
3670         if (IS_ERR(sa))
3671                 return PTR_ERR(sa);
3672
3673         ret = btrfs_scrub_progress(fs_info, sa->devid, &sa->progress);
3674
3675         if (ret == 0 && copy_to_user(arg, sa, sizeof(*sa)))
3676                 ret = -EFAULT;
3677
3678         kfree(sa);
3679         return ret;
3680 }
3681
3682 static long btrfs_ioctl_get_dev_stats(struct btrfs_fs_info *fs_info,
3683                                       void __user *arg)
3684 {
3685         struct btrfs_ioctl_get_dev_stats *sa;
3686         int ret;
3687
3688         sa = memdup_user(arg, sizeof(*sa));
3689         if (IS_ERR(sa))
3690                 return PTR_ERR(sa);
3691
3692         if ((sa->flags & BTRFS_DEV_STATS_RESET) && !capable(CAP_SYS_ADMIN)) {
3693                 kfree(sa);
3694                 return -EPERM;
3695         }
3696
3697         ret = btrfs_get_dev_stats(fs_info, sa);
3698
3699         if (ret == 0 && copy_to_user(arg, sa, sizeof(*sa)))
3700                 ret = -EFAULT;
3701
3702         kfree(sa);
3703         return ret;
3704 }
3705
3706 static long btrfs_ioctl_dev_replace(struct btrfs_fs_info *fs_info,
3707                                     void __user *arg)
3708 {
3709         struct btrfs_ioctl_dev_replace_args *p;
3710         int ret;
3711
3712         if (!capable(CAP_SYS_ADMIN))
3713                 return -EPERM;
3714
3715         p = memdup_user(arg, sizeof(*p));
3716         if (IS_ERR(p))
3717                 return PTR_ERR(p);
3718
3719         switch (p->cmd) {
3720         case BTRFS_IOCTL_DEV_REPLACE_CMD_START:
3721                 if (sb_rdonly(fs_info->sb)) {
3722                         ret = -EROFS;
3723                         goto out;
3724                 }
3725                 if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
3726                         ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3727                 } else {
3728                         ret = btrfs_dev_replace_by_ioctl(fs_info, p);
3729                         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
3730                 }
3731                 break;
3732         case BTRFS_IOCTL_DEV_REPLACE_CMD_STATUS:
3733                 btrfs_dev_replace_status(fs_info, p);
3734                 ret = 0;
3735                 break;
3736         case BTRFS_IOCTL_DEV_REPLACE_CMD_CANCEL:
3737                 p->result = btrfs_dev_replace_cancel(fs_info);
3738                 ret = 0;
3739                 break;
3740         default:
3741                 ret = -EINVAL;
3742                 break;
3743         }
3744
3745         if ((ret == 0 || ret == -ECANCELED) && copy_to_user(arg, p, sizeof(*p)))
3746                 ret = -EFAULT;
3747 out:
3748         kfree(p);
3749         return ret;
3750 }
3751
3752 static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg)
3753 {
3754         int ret = 0;
3755         int i;
3756         u64 rel_ptr;
3757         int size;
3758         struct btrfs_ioctl_ino_path_args *ipa = NULL;
3759         struct inode_fs_paths *ipath = NULL;
3760         struct btrfs_path *path;
3761
3762         if (!capable(CAP_DAC_READ_SEARCH))
3763                 return -EPERM;
3764
3765         path = btrfs_alloc_path();
3766         if (!path) {
3767                 ret = -ENOMEM;
3768                 goto out;
3769         }
3770
3771         ipa = memdup_user(arg, sizeof(*ipa));
3772         if (IS_ERR(ipa)) {
3773                 ret = PTR_ERR(ipa);
3774                 ipa = NULL;
3775                 goto out;
3776         }
3777
3778         size = min_t(u32, ipa->size, 4096);
3779         ipath = init_ipath(size, root, path);
3780         if (IS_ERR(ipath)) {
3781                 ret = PTR_ERR(ipath);
3782                 ipath = NULL;
3783                 goto out;
3784         }
3785
3786         ret = paths_from_inode(ipa->inum, ipath);
3787         if (ret < 0)
3788                 goto out;
3789
3790         for (i = 0; i < ipath->fspath->elem_cnt; ++i) {
3791                 rel_ptr = ipath->fspath->val[i] -
3792                           (u64)(unsigned long)ipath->fspath->val;
3793                 ipath->fspath->val[i] = rel_ptr;
3794         }
3795
3796         ret = copy_to_user((void __user *)(unsigned long)ipa->fspath,
3797                            ipath->fspath, size);
3798         if (ret) {
3799                 ret = -EFAULT;
3800                 goto out;
3801         }
3802
3803 out:
3804         btrfs_free_path(path);
3805         free_ipath(ipath);
3806         kfree(ipa);
3807
3808         return ret;
3809 }
3810
3811 static int build_ino_list(u64 inum, u64 offset, u64 root, void *ctx)
3812 {
3813         struct btrfs_data_container *inodes = ctx;
3814         const size_t c = 3 * sizeof(u64);
3815
3816         if (inodes->bytes_left >= c) {
3817                 inodes->bytes_left -= c;
3818                 inodes->val[inodes->elem_cnt] = inum;
3819                 inodes->val[inodes->elem_cnt + 1] = offset;
3820                 inodes->val[inodes->elem_cnt + 2] = root;
3821                 inodes->elem_cnt += 3;
3822         } else {
3823                 inodes->bytes_missing += c - inodes->bytes_left;
3824                 inodes->bytes_left = 0;
3825                 inodes->elem_missed += 3;
3826         }
3827
3828         return 0;
3829 }
3830
3831 static long btrfs_ioctl_logical_to_ino(struct btrfs_fs_info *fs_info,
3832                                         void __user *arg, int version)
3833 {
3834         int ret = 0;
3835         int size;
3836         struct btrfs_ioctl_logical_ino_args *loi;
3837         struct btrfs_data_container *inodes = NULL;
3838         struct btrfs_path *path = NULL;
3839         bool ignore_offset;
3840
3841         if (!capable(CAP_SYS_ADMIN))
3842                 return -EPERM;
3843
3844         loi = memdup_user(arg, sizeof(*loi));
3845         if (IS_ERR(loi))
3846                 return PTR_ERR(loi);
3847
3848         if (version == 1) {
3849                 ignore_offset = false;
3850                 size = min_t(u32, loi->size, SZ_64K);
3851         } else {
3852                 /* All reserved bits must be 0 for now */
3853                 if (memchr_inv(loi->reserved, 0, sizeof(loi->reserved))) {
3854                         ret = -EINVAL;
3855                         goto out_loi;
3856                 }
3857                 /* Only accept flags we have defined so far */
3858                 if (loi->flags & ~(BTRFS_LOGICAL_INO_ARGS_IGNORE_OFFSET)) {
3859                         ret = -EINVAL;
3860                         goto out_loi;
3861                 }
3862                 ignore_offset = loi->flags & BTRFS_LOGICAL_INO_ARGS_IGNORE_OFFSET;
3863                 size = min_t(u32, loi->size, SZ_16M);
3864         }
3865
3866         path = btrfs_alloc_path();
3867         if (!path) {
3868                 ret = -ENOMEM;
3869                 goto out;
3870         }
3871
3872         inodes = init_data_container(size);
3873         if (IS_ERR(inodes)) {
3874                 ret = PTR_ERR(inodes);
3875                 inodes = NULL;
3876                 goto out;
3877         }
3878
3879         ret = iterate_inodes_from_logical(loi->logical, fs_info, path,
3880                                           build_ino_list, inodes, ignore_offset);
3881         if (ret == -EINVAL)
3882                 ret = -ENOENT;
3883         if (ret < 0)
3884                 goto out;
3885
3886         ret = copy_to_user((void __user *)(unsigned long)loi->inodes, inodes,
3887                            size);
3888         if (ret)
3889                 ret = -EFAULT;
3890
3891 out:
3892         btrfs_free_path(path);
3893         kvfree(inodes);
3894 out_loi:
3895         kfree(loi);
3896
3897         return ret;
3898 }
3899
3900 void btrfs_update_ioctl_balance_args(struct btrfs_fs_info *fs_info,
3901                                struct btrfs_ioctl_balance_args *bargs)
3902 {
3903         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3904
3905         bargs->flags = bctl->flags;
3906
3907         if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags))
3908                 bargs->state |= BTRFS_BALANCE_STATE_RUNNING;
3909         if (atomic_read(&fs_info->balance_pause_req))
3910                 bargs->state |= BTRFS_BALANCE_STATE_PAUSE_REQ;
3911         if (atomic_read(&fs_info->balance_cancel_req))
3912                 bargs->state |= BTRFS_BALANCE_STATE_CANCEL_REQ;
3913
3914         memcpy(&bargs->data, &bctl->data, sizeof(bargs->data));
3915         memcpy(&bargs->meta, &bctl->meta, sizeof(bargs->meta));
3916         memcpy(&bargs->sys, &bctl->sys, sizeof(bargs->sys));
3917
3918         spin_lock(&fs_info->balance_lock);
3919         memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat));
3920         spin_unlock(&fs_info->balance_lock);
3921 }
3922
3923 static long btrfs_ioctl_balance(struct file *file, void __user *arg)
3924 {
3925         struct btrfs_root *root = BTRFS_I(file_inode(file))->root;
3926         struct btrfs_fs_info *fs_info = root->fs_info;
3927         struct btrfs_ioctl_balance_args *bargs;
3928         struct btrfs_balance_control *bctl;
3929         bool need_unlock; /* for mut. excl. ops lock */
3930         int ret;
3931
3932         if (!capable(CAP_SYS_ADMIN))
3933                 return -EPERM;
3934
3935         ret = mnt_want_write_file(file);
3936         if (ret)
3937                 return ret;
3938
3939 again:
3940         if (!test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
3941                 mutex_lock(&fs_info->balance_mutex);
3942                 need_unlock = true;
3943                 goto locked;
3944         }
3945
3946         /*
3947          * mut. excl. ops lock is locked.  Three possibilities:
3948          *   (1) some other op is running
3949          *   (2) balance is running
3950          *   (3) balance is paused -- special case (think resume)
3951          */
3952         mutex_lock(&fs_info->balance_mutex);
3953         if (fs_info->balance_ctl) {
3954                 /* this is either (2) or (3) */
3955                 if (!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
3956                         mutex_unlock(&fs_info->balance_mutex);
3957                         /*
3958                          * Lock released to allow other waiters to continue,
3959                          * we'll reexamine the status again.
3960                          */
3961                         mutex_lock(&fs_info->balance_mutex);
3962
3963                         if (fs_info->balance_ctl &&
3964                             !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
3965                                 /* this is (3) */
3966                                 need_unlock = false;
3967                                 goto locked;
3968                         }
3969
3970                         mutex_unlock(&fs_info->balance_mutex);
3971                         goto again;
3972                 } else {
3973                         /* this is (2) */
3974                         mutex_unlock(&fs_info->balance_mutex);
3975                         ret = -EINPROGRESS;
3976                         goto out;
3977                 }
3978         } else {
3979                 /* this is (1) */
3980                 mutex_unlock(&fs_info->balance_mutex);
3981                 ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS;
3982                 goto out;
3983         }
3984
3985 locked:
3986         BUG_ON(!test_bit(BTRFS_FS_EXCL_OP, &fs_info->flags));
3987
3988         if (arg) {
3989                 bargs = memdup_user(arg, sizeof(*bargs));
3990                 if (IS_ERR(bargs)) {
3991                         ret = PTR_ERR(bargs);
3992                         goto out_unlock;
3993                 }
3994
3995                 if (bargs->flags & BTRFS_BALANCE_RESUME) {
3996                         if (!fs_info->balance_ctl) {
3997                                 ret = -ENOTCONN;
3998                                 goto out_bargs;
3999                         }
4000
4001                         bctl = fs_info->balance_ctl;
4002                         spin_lock(&fs_info->balance_lock);
4003                         bctl->flags |= BTRFS_BALANCE_RESUME;
4004                         spin_unlock(&fs_info->balance_lock);
4005
4006                         goto do_balance;
4007                 }
4008         } else {
4009                 bargs = NULL;
4010         }
4011
4012         if (fs_info->balance_ctl) {
4013                 ret = -EINPROGRESS;
4014                 goto out_bargs;
4015         }
4016
4017         bctl = kzalloc(sizeof(*bctl), GFP_KERNEL);
4018         if (!bctl) {
4019                 ret = -ENOMEM;
4020                 goto out_bargs;
4021         }
4022
4023         if (arg) {
4024                 memcpy(&bctl->data, &bargs->data, sizeof(bctl->data));
4025                 memcpy(&bctl->meta, &bargs->meta, sizeof(bctl->meta));
4026                 memcpy(&bctl->sys, &bargs->sys, sizeof(bctl->sys));
4027
4028                 bctl->flags = bargs->flags;
4029         } else {
4030                 /* balance everything - no filters */
4031                 bctl->flags |= BTRFS_BALANCE_TYPE_MASK;
4032         }
4033
4034         if (bctl->flags & ~(BTRFS_BALANCE_ARGS_MASK | BTRFS_BALANCE_TYPE_MASK)) {
4035                 ret = -EINVAL;
4036                 goto out_bctl;
4037         }
4038
4039 do_balance:
4040         /*
4041          * Ownership of bctl and filesystem flag BTRFS_FS_EXCL_OP goes to
4042          * btrfs_balance.  bctl is freed in reset_balance_state, or, if
4043          * restriper was paused all the way until unmount, in free_fs_info.
4044          * The flag should be cleared after reset_balance_state.
4045          */
4046         need_unlock = false;
4047
4048         ret = btrfs_balance(fs_info, bctl, bargs);
4049         bctl = NULL;
4050
4051         if ((ret == 0 || ret == -ECANCELED) && arg) {
4052                 if (copy_to_user(arg, bargs, sizeof(*bargs)))
4053                         ret = -EFAULT;
4054         }
4055
4056 out_bctl:
4057         kfree(bctl);
4058 out_bargs:
4059         kfree(bargs);
4060 out_unlock:
4061         mutex_unlock(&fs_info->balance_mutex);
4062         if (need_unlock)
4063                 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
4064 out:
4065         mnt_drop_write_file(file);
4066         return ret;
4067 }
4068
4069 static long btrfs_ioctl_balance_ctl(struct btrfs_fs_info *fs_info, int cmd)
4070 {
4071         if (!capable(CAP_SYS_ADMIN))
4072                 return -EPERM;
4073
4074         switch (cmd) {
4075         case BTRFS_BALANCE_CTL_PAUSE:
4076                 return btrfs_pause_balance(fs_info);
4077         case BTRFS_BALANCE_CTL_CANCEL:
4078                 return btrfs_cancel_balance(fs_info);
4079         }
4080
4081         return -EINVAL;
4082 }
4083
4084 static long btrfs_ioctl_balance_progress(struct btrfs_fs_info *fs_info,
4085                                          void __user *arg)
4086 {
4087         struct btrfs_ioctl_balance_args *bargs;
4088         int ret = 0;
4089
4090         if (!capable(CAP_SYS_ADMIN))
4091                 return -EPERM;
4092
4093         mutex_lock(&fs_info->balance_mutex);
4094         if (!fs_info->balance_ctl) {
4095                 ret = -ENOTCONN;
4096                 goto out;
4097         }
4098
4099         bargs = kzalloc(sizeof(*bargs), GFP_KERNEL);
4100         if (!bargs) {
4101                 ret = -ENOMEM;
4102                 goto out;
4103         }
4104
4105         btrfs_update_ioctl_balance_args(fs_info, bargs);
4106
4107         if (copy_to_user(arg, bargs, sizeof(*bargs)))
4108                 ret = -EFAULT;
4109
4110         kfree(bargs);
4111 out:
4112         mutex_unlock(&fs_info->balance_mutex);
4113         return ret;
4114 }
4115
4116 static long btrfs_ioctl_quota_ctl(struct file *file, void __user *arg)
4117 {
4118         struct inode *inode = file_inode(file);
4119         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4120         struct btrfs_ioctl_quota_ctl_args *sa;
4121         int ret;
4122
4123         if (!capable(CAP_SYS_ADMIN))
4124                 return -EPERM;
4125
4126         ret = mnt_want_write_file(file);
4127         if (ret)
4128                 return ret;
4129
4130         sa = memdup_user(arg, sizeof(*sa));
4131         if (IS_ERR(sa)) {
4132                 ret = PTR_ERR(sa);
4133                 goto drop_write;
4134         }
4135
4136         down_write(&fs_info->subvol_sem);
4137
4138         switch (sa->cmd) {
4139         case BTRFS_QUOTA_CTL_ENABLE:
4140                 ret = btrfs_quota_enable(fs_info);
4141                 break;
4142         case BTRFS_QUOTA_CTL_DISABLE:
4143                 ret = btrfs_quota_disable(fs_info);
4144                 break;
4145         default:
4146                 ret = -EINVAL;
4147                 break;
4148         }
4149
4150         kfree(sa);
4151         up_write(&fs_info->subvol_sem);
4152 drop_write:
4153         mnt_drop_write_file(file);
4154         return ret;
4155 }
4156
4157 static long btrfs_ioctl_qgroup_assign(struct file *file, void __user *arg)
4158 {
4159         struct inode *inode = file_inode(file);
4160         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4161         struct btrfs_root *root = BTRFS_I(inode)->root;
4162         struct btrfs_ioctl_qgroup_assign_args *sa;
4163         struct btrfs_trans_handle *trans;
4164         int ret;
4165         int err;
4166
4167         if (!capable(CAP_SYS_ADMIN))
4168                 return -EPERM;
4169
4170         ret = mnt_want_write_file(file);
4171         if (ret)
4172                 return ret;
4173
4174         sa = memdup_user(arg, sizeof(*sa));
4175         if (IS_ERR(sa)) {
4176                 ret = PTR_ERR(sa);
4177                 goto drop_write;
4178         }
4179
4180         trans = btrfs_join_transaction(root);
4181         if (IS_ERR(trans)) {
4182                 ret = PTR_ERR(trans);
4183                 goto out;
4184         }
4185
4186         if (sa->assign) {
4187                 ret = btrfs_add_qgroup_relation(trans, sa->src, sa->dst);
4188         } else {
4189                 ret = btrfs_del_qgroup_relation(trans, sa->src, sa->dst);
4190         }
4191
4192         /* update qgroup status and info */
4193         err = btrfs_run_qgroups(trans);
4194         if (err < 0)
4195                 btrfs_handle_fs_error(fs_info, err,
4196                                       "failed to update qgroup status and info");
4197         err = btrfs_end_transaction(trans);
4198         if (err && !ret)
4199                 ret = err;
4200
4201 out:
4202         kfree(sa);
4203 drop_write:
4204         mnt_drop_write_file(file);
4205         return ret;
4206 }
4207
4208 static long btrfs_ioctl_qgroup_create(struct file *file, void __user *arg)
4209 {
4210         struct inode *inode = file_inode(file);
4211         struct btrfs_root *root = BTRFS_I(inode)->root;
4212         struct btrfs_ioctl_qgroup_create_args *sa;
4213         struct btrfs_trans_handle *trans;
4214         int ret;
4215         int err;
4216
4217         if (!capable(CAP_SYS_ADMIN))
4218                 return -EPERM;
4219
4220         ret = mnt_want_write_file(file);
4221         if (ret)
4222                 return ret;
4223
4224         sa = memdup_user(arg, sizeof(*sa));
4225         if (IS_ERR(sa)) {
4226                 ret = PTR_ERR(sa);
4227                 goto drop_write;
4228         }
4229
4230         if (!sa->qgroupid) {
4231                 ret = -EINVAL;
4232                 goto out;
4233         }
4234
4235         trans = btrfs_join_transaction(root);
4236         if (IS_ERR(trans)) {
4237                 ret = PTR_ERR(trans);
4238                 goto out;
4239         }
4240
4241         if (sa->create) {
4242                 ret = btrfs_create_qgroup(trans, sa->qgroupid);
4243         } else {
4244                 ret = btrfs_remove_qgroup(trans, sa->qgroupid);
4245         }
4246
4247         err = btrfs_end_transaction(trans);
4248         if (err && !ret)
4249                 ret = err;
4250
4251 out:
4252         kfree(sa);
4253 drop_write:
4254         mnt_drop_write_file(file);
4255         return ret;
4256 }
4257
4258 static long btrfs_ioctl_qgroup_limit(struct file *file, void __user *arg)
4259 {
4260         struct inode *inode = file_inode(file);
4261         struct btrfs_root *root = BTRFS_I(inode)->root;
4262         struct btrfs_ioctl_qgroup_limit_args *sa;
4263         struct btrfs_trans_handle *trans;
4264         int ret;
4265         int err;
4266         u64 qgroupid;
4267
4268         if (!capable(CAP_SYS_ADMIN))
4269                 return -EPERM;
4270
4271         ret = mnt_want_write_file(file);
4272         if (ret)
4273                 return ret;
4274
4275         sa = memdup_user(arg, sizeof(*sa));
4276         if (IS_ERR(sa)) {
4277                 ret = PTR_ERR(sa);
4278                 goto drop_write;
4279         }
4280
4281         trans = btrfs_join_transaction(root);
4282         if (IS_ERR(trans)) {
4283                 ret = PTR_ERR(trans);
4284                 goto out;
4285         }
4286
4287         qgroupid = sa->qgroupid;
4288         if (!qgroupid) {
4289                 /* take the current subvol as qgroup */
4290                 qgroupid = root->root_key.objectid;
4291         }
4292
4293         ret = btrfs_limit_qgroup(trans, qgroupid, &sa->lim);
4294
4295         err = btrfs_end_transaction(trans);
4296         if (err && !ret)
4297                 ret = err;
4298
4299 out:
4300         kfree(sa);
4301 drop_write:
4302         mnt_drop_write_file(file);
4303         return ret;
4304 }
4305
4306 static long btrfs_ioctl_quota_rescan(struct file *file, void __user *arg)
4307 {
4308         struct inode *inode = file_inode(file);
4309         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4310         struct btrfs_ioctl_quota_rescan_args *qsa;
4311         int ret;
4312
4313         if (!capable(CAP_SYS_ADMIN))
4314                 return -EPERM;
4315
4316         ret = mnt_want_write_file(file);
4317         if (ret)
4318                 return ret;
4319
4320         qsa = memdup_user(arg, sizeof(*qsa));
4321         if (IS_ERR(qsa)) {
4322                 ret = PTR_ERR(qsa);
4323                 goto drop_write;
4324         }
4325
4326         if (qsa->flags) {
4327                 ret = -EINVAL;
4328                 goto out;
4329         }
4330
4331         ret = btrfs_qgroup_rescan(fs_info);
4332
4333 out:
4334         kfree(qsa);
4335 drop_write:
4336         mnt_drop_write_file(file);
4337         return ret;
4338 }
4339
4340 static long btrfs_ioctl_quota_rescan_status(struct btrfs_fs_info *fs_info,
4341                                                 void __user *arg)
4342 {
4343         struct btrfs_ioctl_quota_rescan_args *qsa;
4344         int ret = 0;
4345
4346         if (!capable(CAP_SYS_ADMIN))
4347                 return -EPERM;
4348
4349         qsa = kzalloc(sizeof(*qsa), GFP_KERNEL);
4350         if (!qsa)
4351                 return -ENOMEM;
4352
4353         if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
4354                 qsa->flags = 1;
4355                 qsa->progress = fs_info->qgroup_rescan_progress.objectid;
4356         }
4357
4358         if (copy_to_user(arg, qsa, sizeof(*qsa)))
4359                 ret = -EFAULT;
4360
4361         kfree(qsa);
4362         return ret;
4363 }
4364
4365 static long btrfs_ioctl_quota_rescan_wait(struct btrfs_fs_info *fs_info,
4366                                                 void __user *arg)
4367 {
4368         if (!capable(CAP_SYS_ADMIN))
4369                 return -EPERM;
4370
4371         return btrfs_qgroup_wait_for_completion(fs_info, true);
4372 }
4373
4374 static long _btrfs_ioctl_set_received_subvol(struct file *file,
4375                                             struct btrfs_ioctl_received_subvol_args *sa)
4376 {
4377         struct inode *inode = file_inode(file);
4378         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4379         struct btrfs_root *root = BTRFS_I(inode)->root;
4380         struct btrfs_root_item *root_item = &root->root_item;
4381         struct btrfs_trans_handle *trans;
4382         struct timespec64 ct = current_time(inode);
4383         int ret = 0;
4384         int received_uuid_changed;
4385
4386         if (!inode_owner_or_capable(inode))
4387                 return -EPERM;
4388
4389         ret = mnt_want_write_file(file);
4390         if (ret < 0)
4391                 return ret;
4392
4393         down_write(&fs_info->subvol_sem);
4394
4395         if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) {
4396                 ret = -EINVAL;
4397                 goto out;
4398         }
4399
4400         if (btrfs_root_readonly(root)) {
4401                 ret = -EROFS;
4402                 goto out;
4403         }
4404
4405         /*
4406          * 1 - root item
4407          * 2 - uuid items (received uuid + subvol uuid)
4408          */
4409         trans = btrfs_start_transaction(root, 3);
4410         if (IS_ERR(trans)) {
4411                 ret = PTR_ERR(trans);
4412                 trans = NULL;
4413                 goto out;
4414         }
4415
4416         sa->rtransid = trans->transid;
4417         sa->rtime.sec = ct.tv_sec;
4418         sa->rtime.nsec = ct.tv_nsec;
4419
4420         received_uuid_changed = memcmp(root_item->received_uuid, sa->uuid,
4421                                        BTRFS_UUID_SIZE);
4422         if (received_uuid_changed &&
4423             !btrfs_is_empty_uuid(root_item->received_uuid)) {
4424                 ret = btrfs_uuid_tree_remove(trans, root_item->received_uuid,
4425                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4426                                           root->root_key.objectid);
4427                 if (ret && ret != -ENOENT) {
4428                         btrfs_abort_transaction(trans, ret);
4429                         btrfs_end_transaction(trans);
4430                         goto out;
4431                 }
4432         }
4433         memcpy(root_item->received_uuid, sa->uuid, BTRFS_UUID_SIZE);
4434         btrfs_set_root_stransid(root_item, sa->stransid);
4435         btrfs_set_root_rtransid(root_item, sa->rtransid);
4436         btrfs_set_stack_timespec_sec(&root_item->stime, sa->stime.sec);
4437         btrfs_set_stack_timespec_nsec(&root_item->stime, sa->stime.nsec);
4438         btrfs_set_stack_timespec_sec(&root_item->rtime, sa->rtime.sec);
4439         btrfs_set_stack_timespec_nsec(&root_item->rtime, sa->rtime.nsec);
4440
4441         ret = btrfs_update_root(trans, fs_info->tree_root,
4442                                 &root->root_key, &root->root_item);
4443         if (ret < 0) {
4444                 btrfs_end_transaction(trans);
4445                 goto out;
4446         }
4447         if (received_uuid_changed && !btrfs_is_empty_uuid(sa->uuid)) {
4448                 ret = btrfs_uuid_tree_add(trans, sa->uuid,
4449                                           BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4450                                           root->root_key.objectid);
4451                 if (ret < 0 && ret != -EEXIST) {
4452                         btrfs_abort_transaction(trans, ret);
4453                         btrfs_end_transaction(trans);
4454                         goto out;
4455                 }
4456         }
4457         ret = btrfs_commit_transaction(trans);
4458 out:
4459         up_write(&fs_info->subvol_sem);
4460         mnt_drop_write_file(file);
4461         return ret;
4462 }
4463
4464 #ifdef CONFIG_64BIT
4465 static long btrfs_ioctl_set_received_subvol_32(struct file *file,
4466                                                 void __user *arg)
4467 {
4468         struct btrfs_ioctl_received_subvol_args_32 *args32 = NULL;
4469         struct btrfs_ioctl_received_subvol_args *args64 = NULL;
4470         int ret = 0;
4471
4472         args32 = memdup_user(arg, sizeof(*args32));
4473         if (IS_ERR(args32))
4474                 return PTR_ERR(args32);
4475
4476         args64 = kmalloc(sizeof(*args64), GFP_KERNEL);
4477         if (!args64) {
4478                 ret = -ENOMEM;
4479                 goto out;
4480         }
4481
4482         memcpy(args64->uuid, args32->uuid, BTRFS_UUID_SIZE);
4483         args64->stransid = args32->stransid;
4484         args64->rtransid = args32->rtransid;
4485         args64->stime.sec = args32->stime.sec;
4486         args64->stime.nsec = args32->stime.nsec;
4487         args64->rtime.sec = args32->rtime.sec;
4488         args64->rtime.nsec = args32->rtime.nsec;
4489         args64->flags = args32->flags;
4490
4491         ret = _btrfs_ioctl_set_received_subvol(file, args64);
4492         if (ret)
4493                 goto out;
4494
4495         memcpy(args32->uuid, args64->uuid, BTRFS_UUID_SIZE);
4496         args32->stransid = args64->stransid;
4497         args32->rtransid = args64->rtransid;
4498         args32->stime.sec = args64->stime.sec;
4499         args32->stime.nsec = args64->stime.nsec;
4500         args32->rtime.sec = args64->rtime.sec;
4501         args32->rtime.nsec = args64->rtime.nsec;
4502         args32->flags = args64->flags;
4503
4504         ret = copy_to_user(arg, args32, sizeof(*args32));
4505         if (ret)
4506                 ret = -EFAULT;
4507
4508 out:
4509         kfree(args32);
4510         kfree(args64);
4511         return ret;
4512 }
4513 #endif
4514
4515 static long btrfs_ioctl_set_received_subvol(struct file *file,
4516                                             void __user *arg)
4517 {
4518         struct btrfs_ioctl_received_subvol_args *sa = NULL;
4519         int ret = 0;
4520
4521         sa = memdup_user(arg, sizeof(*sa));
4522         if (IS_ERR(sa))
4523                 return PTR_ERR(sa);
4524
4525         ret = _btrfs_ioctl_set_received_subvol(file, sa);
4526
4527         if (ret)
4528                 goto out;
4529
4530         ret = copy_to_user(arg, sa, sizeof(*sa));
4531         if (ret)
4532                 ret = -EFAULT;
4533
4534 out:
4535         kfree(sa);
4536         return ret;
4537 }
4538
4539 static int btrfs_ioctl_get_fslabel(struct btrfs_fs_info *fs_info,
4540                                         void __user *arg)
4541 {
4542         size_t len;
4543         int ret;
4544         char label[BTRFS_LABEL_SIZE];
4545
4546         spin_lock(&fs_info->super_lock);
4547         memcpy(label, fs_info->super_copy->label, BTRFS_LABEL_SIZE);
4548         spin_unlock(&fs_info->super_lock);
4549
4550         len = strnlen(label, BTRFS_LABEL_SIZE);
4551
4552         if (len == BTRFS_LABEL_SIZE) {
4553                 btrfs_warn(fs_info,
4554                            "label is too long, return the first %zu bytes",
4555                            --len);
4556         }
4557
4558         ret = copy_to_user(arg, label, len);
4559
4560         return ret ? -EFAULT : 0;
4561 }
4562
4563 static int btrfs_ioctl_set_fslabel(struct file *file, void __user *arg)
4564 {
4565         struct inode *inode = file_inode(file);
4566         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4567         struct btrfs_root *root = BTRFS_I(inode)->root;
4568         struct btrfs_super_block *super_block = fs_info->super_copy;
4569         struct btrfs_trans_handle *trans;
4570         char label[BTRFS_LABEL_SIZE];
4571         int ret;
4572
4573         if (!capable(CAP_SYS_ADMIN))
4574                 return -EPERM;
4575
4576         if (copy_from_user(label, arg, sizeof(label)))
4577                 return -EFAULT;
4578
4579         if (strnlen(label, BTRFS_LABEL_SIZE) == BTRFS_LABEL_SIZE) {
4580                 btrfs_err(fs_info,
4581                           "unable to set label with more than %d bytes",
4582                           BTRFS_LABEL_SIZE - 1);
4583                 return -EINVAL;
4584         }
4585
4586         ret = mnt_want_write_file(file);
4587         if (ret)
4588                 return ret;
4589
4590         trans = btrfs_start_transaction(root, 0);
4591         if (IS_ERR(trans)) {
4592                 ret = PTR_ERR(trans);
4593                 goto out_unlock;
4594         }
4595
4596         spin_lock(&fs_info->super_lock);
4597         strcpy(super_block->label, label);
4598         spin_unlock(&fs_info->super_lock);
4599         ret = btrfs_commit_transaction(trans);
4600
4601 out_unlock:
4602         mnt_drop_write_file(file);
4603         return ret;
4604 }
4605
4606 #define INIT_FEATURE_FLAGS(suffix) \
4607         { .compat_flags = BTRFS_FEATURE_COMPAT_##suffix, \
4608           .compat_ro_flags = BTRFS_FEATURE_COMPAT_RO_##suffix, \
4609           .incompat_flags = BTRFS_FEATURE_INCOMPAT_##suffix }
4610
4611 int btrfs_ioctl_get_supported_features(void __user *arg)
4612 {
4613         static const struct btrfs_ioctl_feature_flags features[3] = {
4614                 INIT_FEATURE_FLAGS(SUPP),
4615                 INIT_FEATURE_FLAGS(SAFE_SET),
4616                 INIT_FEATURE_FLAGS(SAFE_CLEAR)
4617         };
4618
4619         if (copy_to_user(arg, &features, sizeof(features)))
4620                 return -EFAULT;
4621
4622         return 0;
4623 }
4624
4625 static int btrfs_ioctl_get_features(struct btrfs_fs_info *fs_info,
4626                                         void __user *arg)
4627 {
4628         struct btrfs_super_block *super_block = fs_info->super_copy;
4629         struct btrfs_ioctl_feature_flags features;
4630
4631         features.compat_flags = btrfs_super_compat_flags(super_block);
4632         features.compat_ro_flags = btrfs_super_compat_ro_flags(super_block);
4633         features.incompat_flags = btrfs_super_incompat_flags(super_block);
4634
4635         if (copy_to_user(arg, &features, sizeof(features)))
4636                 return -EFAULT;
4637
4638         return 0;
4639 }
4640
4641 static int check_feature_bits(struct btrfs_fs_info *fs_info,
4642                               enum btrfs_feature_set set,
4643                               u64 change_mask, u64 flags, u64 supported_flags,
4644                               u64 safe_set, u64 safe_clear)
4645 {
4646         const char *type = btrfs_feature_set_name(set);
4647         char *names;
4648         u64 disallowed, unsupported;
4649         u64 set_mask = flags & change_mask;
4650         u64 clear_mask = ~flags & change_mask;
4651
4652         unsupported = set_mask & ~supported_flags;
4653         if (unsupported) {
4654                 names = btrfs_printable_features(set, unsupported);
4655                 if (names) {
4656                         btrfs_warn(fs_info,
4657                                    "this kernel does not support the %s feature bit%s",
4658                                    names, strchr(names, ',') ? "s" : "");
4659                         kfree(names);
4660                 } else
4661                         btrfs_warn(fs_info,
4662                                    "this kernel does not support %s bits 0x%llx",
4663                                    type, unsupported);
4664                 return -EOPNOTSUPP;
4665         }
4666
4667         disallowed = set_mask & ~safe_set;
4668         if (disallowed) {
4669                 names = btrfs_printable_features(set, disallowed);
4670                 if (names) {
4671                         btrfs_warn(fs_info,
4672                                    "can't set the %s feature bit%s while mounted",
4673                                    names, strchr(names, ',') ? "s" : "");
4674                         kfree(names);
4675                 } else
4676                         btrfs_warn(fs_info,
4677                                    "can't set %s bits 0x%llx while mounted",
4678                                    type, disallowed);
4679                 return -EPERM;
4680         }
4681
4682         disallowed = clear_mask & ~safe_clear;
4683         if (disallowed) {
4684                 names = btrfs_printable_features(set, disallowed);
4685                 if (names) {
4686                         btrfs_warn(fs_info,
4687                                    "can't clear the %s feature bit%s while mounted",
4688                                    names, strchr(names, ',') ? "s" : "");
4689                         kfree(names);
4690                 } else
4691                         btrfs_warn(fs_info,
4692                                    "can't clear %s bits 0x%llx while mounted",
4693                                    type, disallowed);
4694                 return -EPERM;
4695         }
4696
4697         return 0;
4698 }
4699
4700 #define check_feature(fs_info, change_mask, flags, mask_base)   \
4701 check_feature_bits(fs_info, FEAT_##mask_base, change_mask, flags,       \
4702                    BTRFS_FEATURE_ ## mask_base ## _SUPP,        \
4703                    BTRFS_FEATURE_ ## mask_base ## _SAFE_SET,    \
4704                    BTRFS_FEATURE_ ## mask_base ## _SAFE_CLEAR)
4705
4706 static int btrfs_ioctl_set_features(struct file *file, void __user *arg)
4707 {
4708         struct inode *inode = file_inode(file);
4709         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4710         struct btrfs_root *root = BTRFS_I(inode)->root;
4711         struct btrfs_super_block *super_block = fs_info->super_copy;
4712         struct btrfs_ioctl_feature_flags flags[2];
4713         struct btrfs_trans_handle *trans;
4714         u64 newflags;
4715         int ret;
4716
4717         if (!capable(CAP_SYS_ADMIN))
4718                 return -EPERM;
4719
4720         if (copy_from_user(flags, arg, sizeof(flags)))
4721                 return -EFAULT;
4722
4723         /* Nothing to do */
4724         if (!flags[0].compat_flags && !flags[0].compat_ro_flags &&
4725             !flags[0].incompat_flags)
4726                 return 0;
4727
4728         ret = check_feature(fs_info, flags[0].compat_flags,
4729                             flags[1].compat_flags, COMPAT);
4730         if (ret)
4731                 return ret;
4732
4733         ret = check_feature(fs_info, flags[0].compat_ro_flags,
4734                             flags[1].compat_ro_flags, COMPAT_RO);
4735         if (ret)
4736                 return ret;
4737
4738         ret = check_feature(fs_info, flags[0].incompat_flags,
4739                             flags[1].incompat_flags, INCOMPAT);
4740         if (ret)
4741                 return ret;
4742
4743         ret = mnt_want_write_file(file);
4744         if (ret)
4745                 return ret;
4746
4747         trans = btrfs_start_transaction(root, 0);
4748         if (IS_ERR(trans)) {
4749                 ret = PTR_ERR(trans);
4750                 goto out_drop_write;
4751         }
4752
4753         spin_lock(&fs_info->super_lock);
4754         newflags = btrfs_super_compat_flags(super_block);
4755         newflags |= flags[0].compat_flags & flags[1].compat_flags;
4756         newflags &= ~(flags[0].compat_flags & ~flags[1].compat_flags);
4757         btrfs_set_super_compat_flags(super_block, newflags);
4758
4759         newflags = btrfs_super_compat_ro_flags(super_block);
4760         newflags |= flags[0].compat_ro_flags & flags[1].compat_ro_flags;
4761         newflags &= ~(flags[0].compat_ro_flags & ~flags[1].compat_ro_flags);
4762         btrfs_set_super_compat_ro_flags(super_block, newflags);
4763
4764         newflags = btrfs_super_incompat_flags(super_block);
4765         newflags |= flags[0].incompat_flags & flags[1].incompat_flags;
4766         newflags &= ~(flags[0].incompat_flags & ~flags[1].incompat_flags);
4767         btrfs_set_super_incompat_flags(super_block, newflags);
4768         spin_unlock(&fs_info->super_lock);
4769
4770         ret = btrfs_commit_transaction(trans);
4771 out_drop_write:
4772         mnt_drop_write_file(file);
4773
4774         return ret;
4775 }
4776
4777 static int _btrfs_ioctl_send(struct file *file, void __user *argp, bool compat)
4778 {
4779         struct btrfs_ioctl_send_args *arg;
4780         int ret;
4781
4782         if (compat) {
4783 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
4784                 struct btrfs_ioctl_send_args_32 args32;
4785
4786                 ret = copy_from_user(&args32, argp, sizeof(args32));
4787                 if (ret)
4788                         return -EFAULT;
4789                 arg = kzalloc(sizeof(*arg), GFP_KERNEL);
4790                 if (!arg)
4791                         return -ENOMEM;
4792                 arg->send_fd = args32.send_fd;
4793                 arg->clone_sources_count = args32.clone_sources_count;
4794                 arg->clone_sources = compat_ptr(args32.clone_sources);
4795                 arg->parent_root = args32.parent_root;
4796                 arg->flags = args32.flags;
4797                 memcpy(arg->reserved, args32.reserved,
4798                        sizeof(args32.reserved));
4799 #else
4800                 return -ENOTTY;
4801 #endif
4802         } else {
4803                 arg = memdup_user(argp, sizeof(*arg));
4804                 if (IS_ERR(arg))
4805                         return PTR_ERR(arg);
4806         }
4807         ret = btrfs_ioctl_send(file, arg);
4808         kfree(arg);
4809         return ret;
4810 }
4811
4812 long btrfs_ioctl(struct file *file, unsigned int
4813                 cmd, unsigned long arg)
4814 {
4815         struct inode *inode = file_inode(file);
4816         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4817         struct btrfs_root *root = BTRFS_I(inode)->root;
4818         void __user *argp = (void __user *)arg;
4819
4820         switch (cmd) {
4821         case FS_IOC_GETFLAGS:
4822                 return btrfs_ioctl_getflags(file, argp);
4823         case FS_IOC_SETFLAGS:
4824                 return btrfs_ioctl_setflags(file, argp);
4825         case FS_IOC_GETVERSION:
4826                 return btrfs_ioctl_getversion(file, argp);
4827         case FS_IOC_GETFSLABEL:
4828                 return btrfs_ioctl_get_fslabel(fs_info, argp);
4829         case FS_IOC_SETFSLABEL:
4830                 return btrfs_ioctl_set_fslabel(file, argp);
4831         case FITRIM:
4832                 return btrfs_ioctl_fitrim(fs_info, argp);
4833         case BTRFS_IOC_SNAP_CREATE:
4834                 return btrfs_ioctl_snap_create(file, argp, 0);
4835         case BTRFS_IOC_SNAP_CREATE_V2:
4836                 return btrfs_ioctl_snap_create_v2(file, argp, 0);
4837         case BTRFS_IOC_SUBVOL_CREATE:
4838                 return btrfs_ioctl_snap_create(file, argp, 1);
4839         case BTRFS_IOC_SUBVOL_CREATE_V2:
4840                 return btrfs_ioctl_snap_create_v2(file, argp, 1);
4841         case BTRFS_IOC_SNAP_DESTROY:
4842                 return btrfs_ioctl_snap_destroy(file, argp, false);
4843         case BTRFS_IOC_SNAP_DESTROY_V2:
4844                 return btrfs_ioctl_snap_destroy(file, argp, true);
4845         case BTRFS_IOC_SUBVOL_GETFLAGS:
4846                 return btrfs_ioctl_subvol_getflags(file, argp);
4847         case BTRFS_IOC_SUBVOL_SETFLAGS:
4848                 return btrfs_ioctl_subvol_setflags(file, argp);
4849         case BTRFS_IOC_DEFAULT_SUBVOL:
4850                 return btrfs_ioctl_default_subvol(file, argp);
4851         case BTRFS_IOC_DEFRAG:
4852                 return btrfs_ioctl_defrag(file, NULL);
4853         case BTRFS_IOC_DEFRAG_RANGE:
4854                 return btrfs_ioctl_defrag(file, argp);
4855         case BTRFS_IOC_RESIZE:
4856                 return btrfs_ioctl_resize(file, argp);
4857         case BTRFS_IOC_ADD_DEV:
4858                 return btrfs_ioctl_add_dev(fs_info, argp);
4859         case BTRFS_IOC_RM_DEV:
4860                 return btrfs_ioctl_rm_dev(file, argp);
4861         case BTRFS_IOC_RM_DEV_V2:
4862                 return btrfs_ioctl_rm_dev_v2(file, argp);
4863         case BTRFS_IOC_FS_INFO:
4864                 return btrfs_ioctl_fs_info(fs_info, argp);
4865         case BTRFS_IOC_DEV_INFO:
4866                 return btrfs_ioctl_dev_info(fs_info, argp);
4867         case BTRFS_IOC_BALANCE:
4868                 return btrfs_ioctl_balance(file, NULL);
4869         case BTRFS_IOC_TREE_SEARCH:
4870                 return btrfs_ioctl_tree_search(file, argp);
4871         case BTRFS_IOC_TREE_SEARCH_V2:
4872                 return btrfs_ioctl_tree_search_v2(file, argp);
4873         case BTRFS_IOC_INO_LOOKUP:
4874                 return btrfs_ioctl_ino_lookup(file, argp);
4875         case BTRFS_IOC_INO_PATHS:
4876                 return btrfs_ioctl_ino_to_path(root, argp);
4877         case BTRFS_IOC_LOGICAL_INO:
4878                 return btrfs_ioctl_logical_to_ino(fs_info, argp, 1);
4879         case BTRFS_IOC_LOGICAL_INO_V2:
4880                 return btrfs_ioctl_logical_to_ino(fs_info, argp, 2);
4881         case BTRFS_IOC_SPACE_INFO:
4882                 return btrfs_ioctl_space_info(fs_info, argp);
4883         case BTRFS_IOC_SYNC: {
4884                 int ret;
4885
4886                 ret = btrfs_start_delalloc_roots(fs_info, -1);
4887                 if (ret)
4888                         return ret;
4889                 ret = btrfs_sync_fs(inode->i_sb, 1);
4890                 /*
4891                  * The transaction thread may want to do more work,
4892                  * namely it pokes the cleaner kthread that will start
4893                  * processing uncleaned subvols.
4894                  */
4895                 wake_up_process(fs_info->transaction_kthread);
4896                 return ret;
4897         }
4898         case BTRFS_IOC_START_SYNC:
4899                 return btrfs_ioctl_start_sync(root, argp);
4900         case BTRFS_IOC_WAIT_SYNC:
4901                 return btrfs_ioctl_wait_sync(fs_info, argp);
4902         case BTRFS_IOC_SCRUB:
4903                 return btrfs_ioctl_scrub(file, argp);
4904         case BTRFS_IOC_SCRUB_CANCEL:
4905                 return btrfs_ioctl_scrub_cancel(fs_info);
4906         case BTRFS_IOC_SCRUB_PROGRESS:
4907                 return btrfs_ioctl_scrub_progress(fs_info, argp);
4908         case BTRFS_IOC_BALANCE_V2:
4909                 return btrfs_ioctl_balance(file, argp);
4910         case BTRFS_IOC_BALANCE_CTL:
4911                 return btrfs_ioctl_balance_ctl(fs_info, arg);
4912         case BTRFS_IOC_BALANCE_PROGRESS:
4913                 return btrfs_ioctl_balance_progress(fs_info, argp);
4914         case BTRFS_IOC_SET_RECEIVED_SUBVOL:
4915                 return btrfs_ioctl_set_received_subvol(file, argp);
4916 #ifdef CONFIG_64BIT
4917         case BTRFS_IOC_SET_RECEIVED_SUBVOL_32:
4918                 return btrfs_ioctl_set_received_subvol_32(file, argp);
4919 #endif
4920         case BTRFS_IOC_SEND:
4921                 return _btrfs_ioctl_send(file, argp, false);
4922 #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
4923         case BTRFS_IOC_SEND_32:
4924                 return _btrfs_ioctl_send(file, argp, true);
4925 #endif
4926         case BTRFS_IOC_GET_DEV_STATS:
4927                 return btrfs_ioctl_get_dev_stats(fs_info, argp);
4928         case BTRFS_IOC_QUOTA_CTL:
4929                 return btrfs_ioctl_quota_ctl(file, argp);
4930         case BTRFS_IOC_QGROUP_ASSIGN:
4931                 return btrfs_ioctl_qgroup_assign(file, argp);
4932         case BTRFS_IOC_QGROUP_CREATE:
4933                 return btrfs_ioctl_qgroup_create(file, argp);
4934         case BTRFS_IOC_QGROUP_LIMIT:
4935                 return btrfs_ioctl_qgroup_limit(file, argp);
4936         case BTRFS_IOC_QUOTA_RESCAN:
4937                 return btrfs_ioctl_quota_rescan(file, argp);
4938         case BTRFS_IOC_QUOTA_RESCAN_STATUS:
4939                 return btrfs_ioctl_quota_rescan_status(fs_info, argp);
4940         case BTRFS_IOC_QUOTA_RESCAN_WAIT:
4941                 return btrfs_ioctl_quota_rescan_wait(fs_info, argp);
4942         case BTRFS_IOC_DEV_REPLACE:
4943                 return btrfs_ioctl_dev_replace(fs_info, argp);
4944         case BTRFS_IOC_GET_SUPPORTED_FEATURES:
4945                 return btrfs_ioctl_get_supported_features(argp);
4946         case BTRFS_IOC_GET_FEATURES:
4947                 return btrfs_ioctl_get_features(fs_info, argp);
4948         case BTRFS_IOC_SET_FEATURES:
4949                 return btrfs_ioctl_set_features(file, argp);
4950         case FS_IOC_FSGETXATTR:
4951                 return btrfs_ioctl_fsgetxattr(file, argp);
4952         case FS_IOC_FSSETXATTR:
4953                 return btrfs_ioctl_fssetxattr(file, argp);
4954         case BTRFS_IOC_GET_SUBVOL_INFO:
4955                 return btrfs_ioctl_get_subvol_info(file, argp);
4956         case BTRFS_IOC_GET_SUBVOL_ROOTREF:
4957                 return btrfs_ioctl_get_subvol_rootref(file, argp);
4958         case BTRFS_IOC_INO_LOOKUP_USER:
4959                 return btrfs_ioctl_ino_lookup_user(file, argp);
4960         }
4961
4962         return -ENOTTY;
4963 }
4964
4965 #ifdef CONFIG_COMPAT
4966 long btrfs_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
4967 {
4968         /*
4969          * These all access 32-bit values anyway so no further
4970          * handling is necessary.
4971          */
4972         switch (cmd) {
4973         case FS_IOC32_GETFLAGS:
4974                 cmd = FS_IOC_GETFLAGS;
4975                 break;
4976         case FS_IOC32_SETFLAGS:
4977                 cmd = FS_IOC_SETFLAGS;
4978                 break;
4979         case FS_IOC32_GETVERSION:
4980                 cmd = FS_IOC_GETVERSION;
4981                 break;
4982         }
4983
4984         return btrfs_ioctl(file, cmd, (unsigned long) compat_ptr(arg));
4985 }
4986 #endif